Browse the corpus

Introduction Asthma remains the most common chronic disease of childhood, with 1·1 million children in the UK currently receiving treatment.1 Because of the absence of gold standard tests to confirm or refute asthma, most guidelines concur that asthma is a clinical diagnosis based on a characteristic pattern of symptoms and signs in the absence of an alternative explanation.2 A careful reassessment of a large Canadian cohort of 613 adults with a recent diagnosis of asthma ruled out the diagnosis in a third of those assessed.3 This finding probably represents a combination of asthma remission and overdiagnosis, but in 2% of the population analysed an alternative serious cardiorespiratory disorder was diagnosed. Because of concerns about overdiagnosis (and underdiagnosis), on behalf of the UK National Institute of Health and Care Excellence (NICE), experts have developed comprehensive guidance on the diagnosis of asthma incorporating objective tests, which are yet to be implemented (figure 1).4 Despite the accompanying review highlighting the paucity of evidence in children for the use of lung function measures in the diagnosis of asthma (table 1), the interim report4 proposes a diagnostic algorithm for use in primary care in children with symptoms. The algorithm incorporates the sequential use of four measures of lung function and inflammation, each applied as a dichotomous variable: first, spirometry (forced expiratory volume in 1 s [FEV1] and forced vital capacity [FVC]) expressed as a ratio FEV1:FVC; second, bronchodilator reversibility; third, fractional exhaled nitric oxide (FeNO); and fourth, peak flow variability. For both adults and children, spirometry is the first-line investigation; baseline FEV1 is not included, but rather the FEV1:FVC ratio, with the proposed cutoff for a positive test being an FEV1:FVC ratio of less than 70% or of less than the lower limit of normal if this is known for children. Bronchodilator reversibility is done only if the FEV1:FVC ratio is less than 70%, and deemed positive if FEV1 improves by 12% or more from baseline; children positive to both tests can be diagnosed with asthma.

g an FEV1:FVC ratio of less than 70% or of less than the lower limit of normal if this is known for children. Bronchodilator reversibility is done only if the FEV1:FVC ratio is less than 70%, and deemed positive if FEV1 improves by 12% or more from baseline; children positive to both tests can be diagnosed with asthma. If bronchodilator reversibility is negative, both a FeNO of 35 or more parts per billion and more than 20% variability in peak expiratory flow measured over 2–4 weeks is required to diagnose asthma (ie, three positive tests are needed). For children with normal spirometry (FEV1:FVC ≥70%), bronchodilator reversibility testing is not done, but both a FeNO of 35 or more parts per billion and a more than 20% peak flow variability is required for an asthma diagnosis. Children with other combinations of results should be referred to a specialist for opinion, be reviewed in 6 weeks with repeat tests, or have other diagnoses considered.Figure 1 Current NICE criteria for an asthma diagnosis Criteria to be applied in children with symptoms in keeping with asthma. Peak flow should be monitored for 2–4 weeks. More than 20% variability needed for positive test. For patients in whom asthma is suspected, after 6 weeks repeat abnormal tests and review symptoms. NICE=National Institute for Health and Care Excellence. FEV1:FVC=ratio of forced expiratory volume in 1 s and forced vital capacity. FeNO=fractional exhaled nitric oxide. PEF=peak expiratory flow. ppb=parts per billion. Figure adapted from NICE guidance for children.4

6 weeks repeat abnormal tests and review symptoms. NICE=National Institute for Health and Care Excellence. FEV1:FVC=ratio of forced expiratory volume in 1 s and forced vital capacity. FeNO=fractional exhaled nitric oxide. PEF=peak expiratory flow. ppb=parts per billion. Figure adapted from NICE guidance for children.4 Table 1 Summary of evidence of lung function measures used in the diagnosis of asthma n Sensitivity, % (95% CI) Specificity, % (95% CI) Spirometry FEV1:FVC (no studies identified) ·· NA NA FEV1 <80%5 133 52% (··) 72% (··) Bronchodilator reversibility No studies identified ·· NA NA FeNO >22 ppb6 245 57% (··) 87% (··) Peak flow variability Diurnal peak flow variability (mean over 2 weeks >12·3%)7 61 50% (30–70) 72% (56–84) Studies were restricted to those including only children aged 5–16 years, and the table is adapted from the UK National Institute of Health and Care Excellence interim report.4 We found some additional studies that included mixed populations of adults and children, but because the paediatric data could not be separated out they were not included here.8, 9, 10 FEV1=forced expiratory volume in 1 s. FVC=forced vital capacity. NA=not applicable. FeNO=fractional exhaled nitric oxide. ppb=parts per billion. Research in context Evidence before this study

n Sensitivity, % (95% CI) Specificity, % (95% CI) Spirometry FEV1:FVC (no studies identified) ·· NA NA FEV1 <80%5 133 52% (··) 72% (··) Bronchodilator reversibility No studies identified ·· NA NA FeNO >22 ppb6 245 57% (··) 87% (··) Peak flow variability Diurnal peak flow variability (mean over 2 weeks >12·3%)7 61 50% (30–70) 72% (56–84) Studies were restricted to those including only children aged 5–16 years, and the table is adapted from the UK National Institute of Health and Care Excellence interim report.4 We found some additional studies that included mixed populations of adults and children, but because the paediatric data could not be separated out they were not included here.8, 9, 10 FEV1=forced expiratory volume in 1 s. FVC=forced vital capacity. NA=not applicable. FeNO=fractional exhaled nitric oxide. ppb=parts per billion. Research in context Evidence before this study Asthma is the most common chronic disease in childhood; however, no diagnostic criteria exist to confirm or refute the diagnosis. Guidelines from the British Thoracic Society and Scottish Intercollegiate Guidelines Network recognise that the diagnosis of asthma is a clinical one. Because of concerns about the diagnosis of asthma, particularly overdiagnosis, the UK National Institute of Health and Care Excellence (NICE) has developed comprehensive guidance on the diagnosis incorporating objective tests. The NICE interim report proposes a diagnostic algorithm incorporating the sequential use of four measures of lung function and inflammation in children with symptoms, each applied as a dichotomous variable: spirometry (forced expiratory volume in 1 s [FEV1] and forced vital capacity [FVC] expressed as a ratio FEV1:FVC), bronchodilator reversibility, fractional exhaled nitric oxide, and peak flow variability. For the development of the NICE guideline, systematic scientific literature searches were done to identify all published evidence relevant to the review questions. The authors acknowledged that no evidence was found for asthma diagnosis in children for either FEV1:FVC or bronchodilator reversibility, and little evidence exists for fractional exhaled nitric oxide and peak flow variability. Despite these findings, the guideline proposes a complex algorithm in which at least two of these tests must be positive to make the diagnosis.

for asthma diagnosis in children for either FEV1:FVC or bronchodilator reversibility, and little evidence exists for fractional exhaled nitric oxide and peak flow variability. Despite these findings, the guideline proposes a complex algorithm in which at least two of these tests must be positive to make the diagnosis. Added value of this study To test these proposed algorithms prospectively in newly presenting symptomatic children will take several years. Cohort studies already exist with a wealth of lung function and clinical data and offer an opportunity to quickly assess the likely success of such an algorithm, the sequence, and the proposed cutoffs for the individual tests. In this study, we tested the proposed NICE algorithm using data from a population-based cohort, focusing on adolescents aged 13–16 years with recent asthma symptoms, to simulate the situation in clinical practice. We found poor agreement between the algorithm and our questionnaire-based epidemiological definition of asthma (physician diagnosis, present symptoms, and regular use of inhaled corticosteroids). Implications of all the available evidence

To test these proposed algorithms prospectively in newly presenting symptomatic children will take several years. Cohort studies already exist with a wealth of lung function and clinical data and offer an opportunity to quickly assess the likely success of such an algorithm, the sequence, and the proposed cutoffs for the individual tests. In this study, we tested the proposed NICE algorithm using data from a population-based cohort, focusing on adolescents aged 13–16 years with recent asthma symptoms, to simulate the situation in clinical practice. We found poor agreement between the algorithm and our questionnaire-based epidemiological definition of asthma (physician diagnosis, present symptoms, and regular use of inhaled corticosteroids). Implications of all the available evidence There is wide agreement that an asthma diagnostic pathway incorporating objective measures would be extremely helpful and might subsequently reduce the need for trials of treatment, which can be costly and time consuming. However, evidence so far is clearly insufficient to establish such an algorithm. Our findings challenge the cutoff values defined for spirometry, the order in which the tests are done, and the position of bronchodilator reversibility in the proposed NICE algorithm, which seems to add little in children with symptoms suggestive of asthma. This study makes an important contribution to knowledge in this difficult area. It questions the rationale of the proposed algorithm, which is predominantly based on extrapolated adult data, and highlights the need for an appropriately designed study in the appropriate age group to generate evidence of the value of different tests for the diagnosis of asthma in children.

ge in this difficult area. It questions the rationale of the proposed algorithm, which is predominantly based on extrapolated adult data, and highlights the need for an appropriately designed study in the appropriate age group to generate evidence of the value of different tests for the diagnosis of asthma in children. The accuracy of these tests in this specific sequence for the diagnosis of asthma in children is unknown. Ideally, the proposed algorithms should be tested in prospective studies of newly presenting patients in primary care, which would take several years to complete. Within the Manchester Asthma and Allergy Study, a population-based birth cohort, a wealth of data already exist—including respiratory symptoms, prescribed treatments, and three of the four measures of lung function included in the algorithm—making this a practical setting in which to test a diagnostic algorithm. In this study, we aimed to assess the diagnostic value of each test individually and then test the proposed algorithm using data collected in adolescence, focusing on participants with recent asthma symptoms who were not receiving asthma treatment, to simulate the situation in primary care.

test a diagnostic algorithm. In this study, we aimed to assess the diagnostic value of each test individually and then test the proposed algorithm using data collected in adolescence, focusing on participants with recent asthma symptoms who were not receiving asthma treatment, to simulate the situation in primary care. Methods Study design and population In this analysis, we used data from the Manchester Asthma and Allergy Study,11 a population-based birth cohort study done in two centres in Manchester, UK, using data collected at age 13–16 years. Participants were born in the University Hospital of South Manchester (an academic hospital) and Stepping Hill Hospital (a district general hospital), and the study was done at University Hospital of South Manchester. Participants were recruited prenatally and followed up prospectively at intervals of 2–5 years. The local research ethics committee approved the study. Parents provided written informed consent and children gave assent.11

(a district general hospital), and the study was done at University Hospital of South Manchester. Participants were recruited prenatally and followed up prospectively at intervals of 2–5 years. The local research ethics committee approved the study. Parents provided written informed consent and children gave assent.11 The initial analysis presented uses the whole population, then by subgroup of disease. Additionally, we assumed that children without any respiratory symptoms would not be assessed for asthma diagnosis. Therefore, to simulate what might happen in primary care, we tested the proposed NICE algorithm using follow-up data at age 13–16 years from participants who reported respiratory symptoms (wheeze, cough, or breathlessness) in the previous 12 months and who were not on regular inhaled corticosteroids (because these drugs are associated with improved lung function and decreased FeNO).12 We calculated the proportion of participants with a positive lung function test at each step of the algorithm, and recorded the number of participants that met our definition of current asthma. Data collection At the age 13–16 years follow-up visit, validated questionnaires13 were interviewer-administered in clinic or at home to collect information on parentally reported symptoms, physician-diagnosed diseases, and prescribed treatments. We took spirometry, bronchodilator reversibility, and FeNO measurements.

The initial analysis presented uses the whole population, then by subgroup of disease. Additionally, we assumed that children without any respiratory symptoms would not be assessed for asthma diagnosis. Therefore, to simulate what might happen in primary care, we tested the proposed NICE algorithm using follow-up data at age 13–16 years from participants who reported respiratory symptoms (wheeze, cough, or breathlessness) in the previous 12 months and who were not on regular inhaled corticosteroids (because these drugs are associated with improved lung function and decreased FeNO).12 We calculated the proportion of participants with a positive lung function test at each step of the algorithm, and recorded the number of participants that met our definition of current asthma. Data collection At the age 13–16 years follow-up visit, validated questionnaires13 were interviewer-administered in clinic or at home to collect information on parentally reported symptoms, physician-diagnosed diseases, and prescribed treatments. We took spirometry, bronchodilator reversibility, and FeNO measurements. We defined current asthma as a positive answer to all three of the following questions: “Has the doctor ever told you that your child had asthma?”; “Has your child had wheezing or whistling in the chest in the past 12 months?” and “Has your child had asthma treatment in the past 12 months?” Children whose parents responded negatively to all three questions were defined as non-asthmatic controls. Participants with one or two of the above features, or incomplete data, were defined as having possible asthma. Current rhinitis was defined as a positive response to the question: “In the past 12 months, has your child ever had a problem with sneezing, or a runny nose, or a blocked nose when he/she did not have a cold or flu?”

nts with one or two of the above features, or incomplete data, were defined as having possible asthma. Current rhinitis was defined as a positive response to the question: “In the past 12 months, has your child ever had a problem with sneezing, or a runny nose, or a blocked nose when he/she did not have a cold or flu?” FeNO was measured14 with either a chemiluminescence analyser (NIOX, Solna, Sweden) or an electrochemical analyser (NIOX); chemiluminescence was changed over to an electrochemical analyser on May 4, 2012, with no difference found in recorded results between machines (p=0·74; appendix). Spirometry was measured in the clinic with a pneumotachograph-based system (Jaeger, Wurzberg, Germany) or at home with a flow turbine spirometer (Micro Medical, Basingstoke, UK).15 Data were expressed as percentage predicted FEV116 and FEV1:FVC; we calculated FEV1:FVC predicted values and the lower limit of normal.17, 18 We calculated bronchodilator reversibility as a percentage change after administration of 400 μg of salbutamol with the following equation: (post-bronchodilator FEV1-baseline FEV1)×100baseline FEV1 Bronchodilator reversibility was deemed positive if FEV1 increased by 12% or more. Peak flow variability was not measured.

FeNO was measured14 with either a chemiluminescence analyser (NIOX, Solna, Sweden) or an electrochemical analyser (NIOX); chemiluminescence was changed over to an electrochemical analyser on May 4, 2012, with no difference found in recorded results between machines (p=0·74; appendix). Spirometry was measured in the clinic with a pneumotachograph-based system (Jaeger, Wurzberg, Germany) or at home with a flow turbine spirometer (Micro Medical, Basingstoke, UK).15 Data were expressed as percentage predicted FEV116 and FEV1:FVC; we calculated FEV1:FVC predicted values and the lower limit of normal.17, 18 We calculated bronchodilator reversibility as a percentage change after administration of 400 μg of salbutamol with the following equation: (post-bronchodilator FEV1-baseline FEV1)×100baseline FEV1 Bronchodilator reversibility was deemed positive if FEV1 increased by 12% or more. Peak flow variability was not measured. Statistical analysis We calculated descriptive statistics for diagnostic tests and made comparisons with χ2 tests, one-way ANOVA, and Fisher's exact test, as appropriate. Among study participants who met the definition of current asthma or who were non-asthmatic (excluding participants with possible asthma), the variables in the algorithm were assessed with sensitivities, specificities, positive predictive values, negative predictive values, and areas under receiver-operating characteristic curves (AUROCs). We used a multivariable logistic regression model, assuming a linear functional form for the predictors, to investigate the importance of the considered variables. We used Youden's J statistic to estimate the best cutoff values of each test individually. We did all analyses with SPSS version 22 and used a 5% significance level throughout.

ble logistic regression model, assuming a linear functional form for the predictors, to investigate the importance of the considered variables. We used Youden's J statistic to estimate the best cutoff values of each test individually. We did all analyses with SPSS version 22 and used a 5% significance level throughout. Role of the funding source The funders and sponsors of the study had no role in study design, data collection, data analysis, data interpretation, or writing of the report. The corresponding author had full access to all the data in the study and had final responsibility for the decision to submit for publication.

ble logistic regression model, assuming a linear functional form for the predictors, to investigate the importance of the considered variables. We used Youden's J statistic to estimate the best cutoff values of each test individually. We did all analyses with SPSS version 22 and used a 5% significance level throughout. Role of the funding source The funders and sponsors of the study had no role in study design, data collection, data analysis, data interpretation, or writing of the report. The corresponding author had full access to all the data in the study and had final responsibility for the decision to submit for publication. Results Of the 1184 children born into the cohort, 772 attended follow-up at age 13–16 years (mean 15·5 years [SD 0·64]) between July 22, 2011, and Nov 11, 2014 (appendix). Characteristics of the study population are presented in table 2. Comparisons between children included and children excluded in the study, and between male and female participants, are presented in the appendix. Among the 772 participants reviewed, 630 (82%) had measurements for spirometry, 624 (81%) for bronchodilator reversibility, and 485 (63%) for FeNO (table 2); 481 (62%) had all three measurements of lung function, 189 (39%) of whom reported one or more symptom in keeping with possible asthma in the past 12 months. Further descriptions of the lung function parameters are presented in the appendix. For the whole population, measured mean FEV1:FVC for girls (89·7% [95% CI 89·0–90·4]) and for boys (86·9% [86·1–87·7]) were very similar to the respective predicted values (89·5% vs 86·3%; appendix). Only ten (2%) of 630 children had an FEV1:FVC of less than 70% (two with asthma). The mean calculated lower limit of normal for FEV1:FVC for girls was 78·2% (95% CI 78·2–78·3) and for boys was 74·8% (74·8–74·9; appendix); 28 (4%) of 630 children had an FEV1:FVC below the lower limit of normal, of whom 11 had asthma. An increase after bronchodilator use of FEV1 of 12% or more from baseline was seen in 54 (9%) of 624 children (table 2). FeNO was 35 or more parts per billion in 115 (24%) of 485 children, of whom 29 had asthma.Table 2 Characteristics of the study population

1:FVC below the lower limit of normal, of whom 11 had asthma. An increase after bronchodilator use of FEV1 of 12% or more from baseline was seen in 54 (9%) of 624 children (table 2). FeNO was 35 or more parts per billion in 115 (24%) of 485 children, of whom 29 had asthma.Table 2 Characteristics of the study population Whole population Non-asthmatic individuals Individuals with possible asthma Individuals with current asthma p value* Spirometry Number of patients 630 403 153 74 NA Sex Male 325 (52%) 193 (48%) 91 (59%) 41 (55%) 0·040† Female 305 (48%) 210 (52%) 62 (41%) 33 (45%) NA Age (years) 15·6 (15·5–15·6) 15·6 (15·6–15·7) 15·4 (15·3–15·6) 15·4 (15·2–15·6) 0·0021‡ FEV1 (L) 3·6 (3·6–3·7) 3·7 (3·6–3·7) 3·7 (3·6–3·9) 3·3 (3·1–3·5) 0·00012‡ FEV1 (% predicted) 98·7% (97·7–99·7) 99·3% (98·2–100·4) 100·0% (97·9–102·1) 92·9% (89·6–96·2) <0·0001‡ FEV1:FVC 88·3% (87·7–88·8) 89·2% (88·5–89·8) 87·9% (86·9–89·0) 84·0% (82·3–85·9) <0·0001‡ FEV1:FVC <70% 10 (2%) 5 (1%) 3 (2%) 2 (3%) 0·44§ Currently on regular inhaled corticosteroids 34 (5%) 0 5 (3%) 29 (39%) ·· Bronchodilator reversibility Number of patients 624 399 151 74 NA Bronchodilator reversibility 4·9% (4·4–5·3) 4·5% (3·9–5·0) 4·8% (4·0–5·7) 7·2% (5·7–8·8) 0·00043‡ Bronchodilator reversibility ≥12% 54 (9%) 26 (7%) 16 (11%) 12 (16%) 0·015† FeNO Number of patients 485 314 115 56 NA Geometric mean FeNO (ppb) 20·0 (18·8–21·6) 17·4 (16·2–18·8) 22·2 (19·1–25·8) 35·7 (27·3–46·6) <0·0001‡ FeNO ≥35 ppb 115 (24%) 54 (17%) 32 (28%) 29 (52%) <0·0001† Data are n (%) or mean (95% CI), unless stated otherwise. NA=not applicable. FEV1=forced expiratory volume in 1 s. FVC=forced vital capacity. FeNO=fractional exhaled nitric oxide. ppb=parts per billion.

7·4 (16·2–18·8) 22·2 (19·1–25·8) 35·7 (27·3–46·6) <0·0001‡ FeNO ≥35 ppb 115 (24%) 54 (17%) 32 (28%) 29 (52%) <0·0001† Data are n (%) or mean (95% CI), unless stated otherwise. NA=not applicable. FEV1=forced expiratory volume in 1 s. FVC=forced vital capacity. FeNO=fractional exhaled nitric oxide. ppb=parts per billion. * p values compare the three disease groups. † χ2 test. ‡ One-way ANOVA. § Fisher's exact test. We assessed the diagnostic value of each test in children with current asthma (n=74) and without asthma (n=403), excluding children with possible asthma (n=153); for completeness, we included % predicted FEV1 (table 3). Our results suggest that values less stringent than those proposed in the NICE algorithm were much more informative (FEV1:FVC <83·8%, bronchodilator reversibility ≥3·48%, and FeNO ≥24 parts per billion; table 3).Table 3 Diagnostic values for individual tests in 477 children

ness, we included % predicted FEV1 (table 3). Our results suggest that values less stringent than those proposed in the NICE algorithm were much more informative (FEV1:FVC <83·8%, bronchodilator reversibility ≥3·48%, and FeNO ≥24 parts per billion; table 3).Table 3 Diagnostic values for individual tests in 477 children Sensitivity Specificity Positive predictive value Negative predictive value FEV1:FVC (n=477; AUROC=0·701) <70% 2/74 (3%) 398/403 (99%) 2/7 (29%) 398/470 (85%) <75% 8/74 (11%) 391/403 (97%) 8/20 (40%) 391/457 (86%) <80% 20/74 (27%) 368/403 (91%) 20/55 (36%) 368/422 (87%) LLN 11/74 (15%) 391/403 (97%) 11/23 (48%) 391/454 (86%) <83·8%* 40/74 (54%) 328/403 (81%) 40/115 (35%) 328/362 (91%) Bronchodilator reversibility (n=473; AUROC=0·636) ≥12% 12/74 (16%) 373/399 (93%) 12/38 (32%) 373/435 (86%) ≥15% 7/74 (9%) 380/399 (95%) 7/26 (27%) 380/447 (85%) ≥3·48%* 57/74 (77%) 181/399 (45%) 57/275 (21%) 181/198 (91%) FeNO (n=370; AUROC=0·711) ≥35 ppb 29/56 (52%) 260/314 (83%) 29/83 (35%) 260/287 (91%) ≥40 ppb 26/56 (46%) 272/314 (87%) 26/68 (38%) 272/302 (90%) ≥24 ppb* 35/56 (63%) 228/314 (73%) 35/121 (29%) 228/249 (92%) FEV1 % predicted†(n=477; AUROC=0·623) <80% 16/74 (22%) 384/403 (95%) 16/35 (46%) 384/442 (87%) <91·045%* 32/74 (43%) 312/403 (77%) 32/123 (26%) 312/354 (88%) Data include children with asthma (n=74) and non-asthmatic controls (n=403) who had spirometry. Children with possible asthma were excluded. FEV1=forced expiratory volume in 1 s. FVC=forced vital capacity. AUROC=area under the receiver-operating characteristic curve. LLN=lower limit of normal. FeNO=fractional exhaled nitric oxide. ppb=parts per billion.

n=74) and non-asthmatic controls (n=403) who had spirometry. Children with possible asthma were excluded. FEV1=forced expiratory volume in 1 s. FVC=forced vital capacity. AUROC=area under the receiver-operating characteristic curve. LLN=lower limit of normal. FeNO=fractional exhaled nitric oxide. ppb=parts per billion. * The values denote the best cutoffs according to Youden's J statistic (sensitivity + specificity − 1) for this population. † FEV1 % predicted does not form part of the diagnostic algorithm but has been included for completeness. Multivariable logistic regression models revealed that FeNO (p<0·0001) and FEV1:FVC (p=0·0075), but not bronchodilator reversibility (p=0·97), were independently associated with asthma (appendix). Each unit increase in FeNO was associated with an odds ratio of 1·03 (95% CI 1·02–1·04) for asthma, whereas each 1% decrease in FEV1:FVC was associated with an odds ratio of 1·10 (1·04–1·16). This model had an AUROC of 0·79 (95% CI 0·72–0·86) for predicting asthma, which is higher than the AUROC values for the individual tests (appendix). We found no difference in the magnitude of the effect for FeNO and FEV1:FVC when bronchodilator reversibility was removed from the model (appendix).

o of 1·10 (1·04–1·16). This model had an AUROC of 0·79 (95% CI 0·72–0·86) for predicting asthma, which is higher than the AUROC values for the individual tests (appendix). We found no difference in the magnitude of the effect for FeNO and FEV1:FVC when bronchodilator reversibility was removed from the model (appendix). Of the 481 children with full lung function data, 56 (12%) had current asthma and 310 (64%) did not have asthma (figure 2); 115 (24%) with possible asthma were excluded. Only six children (four with asthma) had positive results for all three tests (spirometry, bronchodilator reversibility, and FeNO). Conversely, 24 (43%) of the 56 children with asthma were negative on all three tests.Figure 2 Venn diagrams of overlap between positive tests for spirometry, bronchodilator reversibility, and FeNO in different subgroups of children

sitive results for all three tests (spirometry, bronchodilator reversibility, and FeNO). Conversely, 24 (43%) of the 56 children with asthma were negative on all three tests.Figure 2 Venn diagrams of overlap between positive tests for spirometry, bronchodilator reversibility, and FeNO in different subgroups of children (A,C) Data are number of patients with positive tests (number of patients with current asthma; percentage with asthma). Patients within the orange circles have a FeNO of 35 or more parts per billion, patients within the green circles have an FEV1:FVC of less than the lower limit of normal, and patients within the purple circles have bronchodilator reversibility of 12% or higher. (A) Children with complete data (n=366; 56 with current asthma and 310 without asthma; children with possible asthma excluded; 259 all negative tests [24, 9%] with current asthma). (B) Children with current asthma (n=56; 310 without asthma and children with possible asthma excluded; 24 all negative tests). (C) Children with recent symptoms but not on regular ICS (n=89; 34 with current asthma and 55 without asthma; children with possible asthma excluded; 58 all negative tests [17, 29%] with current asthma). (D) Children with recent symptoms and current asthma who are not on ICS (n=34; 17 all negative tests). FEV1:FVC=ratio of forced expiratory volume in 1 s and forced vital capacity. FeNO=fractional exhaled nitric oxide. ICS=inhaled corticosteroids.

le asthma excluded; 58 all negative tests [17, 29%] with current asthma). (D) Children with recent symptoms and current asthma who are not on ICS (n=34; 17 all negative tests). FEV1:FVC=ratio of forced expiratory volume in 1 s and forced vital capacity. FeNO=fractional exhaled nitric oxide. ICS=inhaled corticosteroids. A total of 189 children reported one or more respiratory symptoms within the previous 12 months, of whom 26 received regular inhaled corticosteroids and were excluded from this analysis. Of the remaining 163 children, 34 had current asthma, 55 were classified as non-asthmatic controls (reported cough or breathlessness, but not wheeze), and 74 had possible asthma (excluded from this analysis). The predictive values of each test in this population are shown in table 4 and the appendix.Table 4 Diagnostic values for individual tests in 163 children reporting one or more respiratory symptoms

tic controls (reported cough or breathlessness, but not wheeze), and 74 had possible asthma (excluded from this analysis). The predictive values of each test in this population are shown in table 4 and the appendix.Table 4 Diagnostic values for individual tests in 163 children reporting one or more respiratory symptoms Sensitivity Specificity Positive predictive value Negative predictive value FEV1:FVC (n=89; AUROC=0·616) <70% 0/34 (0%) 53/55 (96%) 0/2 (0%) 53/87 (61%) <75% 2/34 (6%) 51/55 (93%) 2/6 (33%) 51/83 (61%) <80% 8/34 (24%) 48/55 (87%) 8/15 (53%) 48/74 (65%) LLN 4/34 (12%) 51/55 (93%) 4/8 (50%) 51/81 (63%) <85·5%* 19/34 (56%) 38/55 (69%) 19/36 (53%) 38/53 (72%) Bronchodilator reversibility (n=89; AUROC=0·594) ≥12% 3/34 (9%) 51/55 (93%) 3/7 (43%) 51/82 (62%) ≥15% 2/34 (6%) 52/55 (95%) 2/5 (40%) 52/84 (62%) ≥3·2%* 27/34 (79%) 23/55 (42%) 27/59 (46%) 23/30 (77%) FeNO (n=89; AUROC=0·618) ≥35 ppb 15/34 (44%) 46/55 (84%) 15/24 (63%) 46/65 (71%) ≥40 ppb 13/34 (38%) 49/55 (89%) 13/19 (68%) 49/70 (70%) ≥37 ppb* 15/34 (44%) 47/55 (85%) 15/23 (65%) 47/66 (71%) FEV1 % predicted†(n=89; AUROC=0·555) <80% 3/34 (9%) 51/55 (93%) 3/7 (43%) 51/82 (62%) <106·5%* 27/34 (79%) 18/55 (33%) 27/64 (42%) 18/25 (72%) Data include symptomatic children not using regular inhaled corticosteroids with asthma (n=34) and non-asthmatic controls (n=55). Children with possible asthma were excluded. FEV1=forced expiratory volume in 1 s. FVC=forced vital capacity. AUROC=area under the receiver-operating characteristic curve. LLN=lower limit of normal. FeNO=fractional exhaled nitric oxide. ppb=parts per billion.

eroids with asthma (n=34) and non-asthmatic controls (n=55). Children with possible asthma were excluded. FEV1=forced expiratory volume in 1 s. FVC=forced vital capacity. AUROC=area under the receiver-operating characteristic curve. LLN=lower limit of normal. FeNO=fractional exhaled nitric oxide. ppb=parts per billion. * These values are the best cutoffs according to Youden's J statistic (sensitivity + specificity − 1). † FEV1 % predicted does not form part of the diagnostic algorithm but has been included for completeness. In the multivariable logistic regression analysis (appendix), the only independent predictor of asthma was FeNO; for each unit increase in FeNO, there was an odds ratio for asthma of 1·02 (95% CI 1·01–1·04; p=0·0062). This model had an AUROC of 0·68 (95% CI 0·57–0·80) for predicting asthma. When bronchodilator reversibility was removed, FeNO remained the only independent predictor of asthma, with a non-significant trend for FEV1:FVC (p=0·096; appendix). Among 89 children with recent symptoms but not on regular inhaled corticosteroids (34 with asthma and 55 without asthma), only two children were positive for all three tests, both of whom had asthma (figure 2). Of 21 participants in this population with only a positive test for FeNO, 13 (62%) had asthma. 17 (50%) of 34 children with asthma and not on regular inhaled corticosteroids were negative on all three tests (figure 2).

out asthma), only two children were positive for all three tests, both of whom had asthma (figure 2). Of 21 participants in this population with only a positive test for FeNO, 13 (62%) had asthma. 17 (50%) of 34 children with asthma and not on regular inhaled corticosteroids were negative on all three tests (figure 2). We passed these 89 children with symptoms but not on regular inhaled corticosteroids (34 with current asthma, 55 without asthma) through the sequential NICE diagnostic algorithm, stopping when a diagnosis was reached or the next test (peak flow variability) was unavailable (figure 3). Only two children had an FEV1:FVC of less than 70% (figure 3); both had bronchodilator reversibility of 12% or higher, meeting the algorithm criteria for asthma. Neither of these children had current asthma, and FeNO was less than 35 parts per billion in both cases. Of the remaining 87 children with an FEV1:FVC of 70% or higher, 24 had a FeNO of 35 or more parts per billion and would be diagnosed with asthma or suspected asthma, depending on the results of peak expiratory flow monitoring. Of these 24 children, 15 had current asthma and nine were non-asthmatics. FeNO was less than 35 parts per billion in the remaining 63 children (19 with current asthma). If peak expiratory flow diaries had shown 20% reversibility for 4 days over 2 weeks in these children, they would fall into the suspect asthma part of the algorithm, in which case tests should be repeated at 6 weeks. For completeness, we repeated this analysis with an FEV1:FVC of less than the lower limit of normal (figure 3, appendix) and for all symptomatic children (n=163), including those with possible asthma (appendix).Figure 3 Diagnostic algorithm for 89 children with current symptoms not on inhaled corticosteroids

t 6 weeks. For completeness, we repeated this analysis with an FEV1:FVC of less than the lower limit of normal (figure 3, appendix) and for all symptomatic children (n=163), including those with possible asthma (appendix).Figure 3 Diagnostic algorithm for 89 children with current symptoms not on inhaled corticosteroids The number in parenthesis denotes the number of children with this test result who had asthma, according to our questionnaire-based definition. PEF data were not available in this population. Obstructive spirometry denoted with an FEV1:FVC of less than 70% (A) and of less than the lower limit of normal (B). FEV1:FVC=ratio of forced expiratory volume in 1 s and forced vital capacity. FeNO=fractional exhaled nitric oxide. PEF=peak expiratory flow. ppb=parts per billion. Discussion Using data from a population-based birth cohort, we tested the NICE asthma diagnostic algorithm based on four measures of lung function, which has been proposed for use in primary care. We found poor agreement between the algorithm and our questionnaire-based epidemiological definition of current asthma (physician diagnosis, current symptoms, and current medication) in adolescents aged 13–16 years. In particular, our findings challenge the cutoff values defined for spirometry, the order in which the tests are done, and the position of bronchodilator reversibility within the algorithm sequence, which seems to add little when tested in this population.

ptoms, and current medication) in adolescents aged 13–16 years. In particular, our findings challenge the cutoff values defined for spirometry, the order in which the tests are done, and the position of bronchodilator reversibility within the algorithm sequence, which seems to add little when tested in this population. Among symptomatic children who had completed three of the four tests, we were able to securely diagnose asthma in only two children using the algorithm, neither of whom met our epidemiological definition; this number increased to five if we used the lower limit of normal rather than less than 70% as the cutoff for FEV1:FVC. Almost all other children would require an additional 2 weeks of peak flow monitoring for asthma to be diagnosed or suspected. The economic evaluation of peak flow variability assumes a manual diary will be returned to a practice nurse who will do the calculation (>20% variability on ≥3 days) within 10 min.4 Results from a previous study19 showed large discrepancies between values recorded electronically and those transcribed into a diary. For peak flow variability to be useful, an electronic peak flow meter (which calculates the variability) will probably be required, changing the economic model.

≥3 days) within 10 min.4 Results from a previous study19 showed large discrepancies between values recorded electronically and those transcribed into a diary. For peak flow variability to be useful, an electronic peak flow meter (which calculates the variability) will probably be required, changing the economic model. The proposed algorithm always starts with an FEV1:FVC assessment, with a cutoff of 70% (or lower limit of normal in children if available) required to diagnose obstruction. Within our population-based sample, a finding of an FEV1:FVC of less than 70% was rare, occurring in only ten (2%) of 630 children, most of whom did not have current asthma. Although we identified additional children with obstructed spirometry (28 [4%] of 630) by replacing the FEV1:FVC cutoff of less than 70% with a cutoff of less than the lower limit of normal, we found that the best cutoff for FEV1:FVC was much higher at 85·5%, which produced a sensitivity of 56% (19 of 34 participants; compared with a sensitivity of 0% [0 of 34] for FEV1:FVC <70%). The lower limit of normal for FEV1:FVC falls with age and differs between the sexes, only reaching 70% for most adults by around age 50 years, suggesting that a 70% cutoff is too low for children (appendix).17 We recognise that it would not be practical to suggest different cutoff values for boys and girls for each age within a diagnostic algorithm, unless it was fully computerised. In the interim, it would seem sensible to place greater emphasis on the use of the lower limit of normal value, and recommend that this value is made available on spirometers used in primary care. We note that the national guidelines of some countries recommend much higher FEV1:FVC values to indicate obstruction (eg, Canada recommends an FEV1:FVC of 80%).20 Use of FEV1:FVC as the initial screen for adults is necessary because the differential diagnosis includes chronic obstructive pulmonary disease and interstitial lung disease, but these diseases are unlikely to be relevant in children.

FEV1:FVC values to indicate obstruction (eg, Canada recommends an FEV1:FVC of 80%).20 Use of FEV1:FVC as the initial screen for adults is necessary because the differential diagnosis includes chronic obstructive pulmonary disease and interstitial lung disease, but these diseases are unlikely to be relevant in children. For those with obstructed spirometry, the next test in the algorithm is bronchodilator reversibility. In our population-based sample, bronchodilator reversibility was seen in 54 (9%) of 624 children, most of whom did not have obstructive spirometry. However, following the diagnostic algorithm in children with symptoms, only eight children would proceed to bronchodilator reversibility (five with positive results, of whom three had current asthma). In children with symptoms of asthma, a cutoff of 12% had only 9% (three of 34 participants) sensitivity for current asthma; the highest sensitivity (79·4%; 27 of 34) was seen for a cutoff of 3·2% (lower than the mean value in this population). Bronchodilator reversibility was not significantly associated with current asthma in the multivariable analysis, perhaps reflecting the negative correlation with FEV1:FVC (appendix). Overall, bronchodilator reversibility was less informative than other tests for the diagnosis of current asthma.

lue in this population). Bronchodilator reversibility was not significantly associated with current asthma in the multivariable analysis, perhaps reflecting the negative correlation with FEV1:FVC (appendix). Overall, bronchodilator reversibility was less informative than other tests for the diagnosis of current asthma. FeNO was positive (≥35 parts per billion) in almost a quarter of all children and was the most sensitive test (44% [15 of 34 children]) among children with symptoms. Results from a previous study6 in symptomatic children suggested a cutoff of 22 parts per billion, when asthma was diagnosed on the basis of a positive methacholine challenge or bronchodilator reversibility; our data suggested the best cutoff was 37 parts per billion. When we followed the proposed algorithm, FeNO needed to be measured in more than 90% of children. Lung function guidelines of the American Thoracic Society and European Respiratory Society14 recommend FeNO be measured before spirometry in adults because spirometry can reduce FeNO by up to 25% in patients both with and without asthma.21, 22 Considering that FeNO has the best diagnostic accuracy for current asthma, that within the present algorithm almost all children needed FeNO measurement, and that unless done first the value might be an underestimate, we think that measuring FeNO first in children would seem logical.

both with and without asthma.21, 22 Considering that FeNO has the best diagnostic accuracy for current asthma, that within the present algorithm almost all children needed FeNO measurement, and that unless done first the value might be an underestimate, we think that measuring FeNO first in children would seem logical. The ideal setting in which to test the proposed diagnostic algorithm would be a prospective study of patients with newly presenting symptoms who undergo lung function testing while acutely symptomatic before the introduction of any treatment. The outcome of such a study is unlikely to be known within 5 years. Therefore, recognising the paucity of population-based paediatric lung function data in the public domain and the opportunity to test the algorithm in children with both lung function measurements and parent-reported symptom data, we did this assessment using data from our birth cohort. Our study relied on a physician diagnosis of asthma, without details of how this was ascertained. However, we believe that physician diagnosis, together with information on ongoing symptoms and prescribed drugs, is as robust a definition of current asthma that we could achieve in epidemiology. We also took the measurements of lung function at routine visits, without specifically waiting until the child was acutely symptomatic. However, in clinical practice, the general practitioner is unlikely to perform lung function tests when a patient presents with acute symptoms during the consultation. Additionally, because peak flow variability had not been measured prospectively in this birth cohort, we had only data for three of the four possible tests available to analyse. We note that maternal smoking in pregnancy was more common in participants who did not attend follow-up, and it remains possible, although unlikely, that inclusion of such children would have improved the performance of the algorithm. We acknowledge that some of the children who were included as non-asthmatic controls could have had asthma at a younger age. Therefore, we repeated the analysis with these 13 children excluded and found the results were not materially different (data available on request).

ave improved the performance of the algorithm. We acknowledge that some of the children who were included as non-asthmatic controls could have had asthma at a younger age. Therefore, we repeated the analysis with these 13 children excluded and found the results were not materially different (data available on request). Asthma is an umbrella term that includes several phenotypes (which are better characterised in adults than in children); however, at present, the diagnosis is made before any phenotyping and treatment pathways are not dependent on phenotyping for most children with mild-to-moderate asthma. Diagnostic algorithms might need to allow for different patterns of lung function to capture all types of asthma. We recognise that we have tested this algorithm in a population of adolescents (aged 13–16 years) and that this age range is not fully representative of the paediatric population. Results in younger children might reveal different strengths and weaknesses of this algorithm and future prospective studies should aim to include children from age 5–6 years upwards. We believe that our study makes an important contribution to knowledge in this difficult area. Our findings suggest an urgent need exists for appropriately designed studies to generate evidence about the value of different tests for the diagnosis of paediatric asthma in primary care. Until such evidence is available, the proposed NICE guidance on asthma diagnosis should not be implemented in children. Supplementary Material Supplementary appendix

We believe that our study makes an important contribution to knowledge in this difficult area. Our findings suggest an urgent need exists for appropriately designed studies to generate evidence about the value of different tests for the diagnosis of paediatric asthma in primary care. Until such evidence is available, the proposed NICE guidance on asthma diagnosis should not be implemented in children. Supplementary Material Supplementary appendix Acknowledgments We thank the children and their parents for their continued support and enthusiasm. We greatly appreciate their commitment to the project. We also acknowledge the hard work and dedication of the study team (post-doctoral scientists, research fellows, nurses, physiologists, technicians, and clerical staff). The Manchester Asthma and Allergy Study was supported by Asthma UK grants number 301 (1995–98), number 362 (1998–2001), number 01/012 (2001–04), and number 04/014 (2004–07); the BMA James Trust (2005); the JP Moulton Charitable Foundation (2004 to present); the North West Lung Centre Charity (1997 to present); and UK Medical Research Council grants G0601361 (2007–2012), MR/K002449/1 (2013–2014), and MR/L012693/1 (2014–2018). HD is supported by an Asthma UK Senior Clinical Academic Development Award. This report is independent research supported by UK National Institute for Health Research South Manchester Respiratory and Allergy Clinical Research Facility at University Hospital of South Manchester NHS Foundation Trust. The views expressed in this publication are those of the authors and not necessarily those of the UK National Health Service, the UK National Institute for Health Research, or the UK Department of Health.

Respiratory and Allergy Clinical Research Facility at University Hospital of South Manchester NHS Foundation Trust. The views expressed in this publication are those of the authors and not necessarily those of the UK National Health Service, the UK National Institute for Health Research, or the UK Department of Health. Contributors CM and AS contributed to the concept and design of the study, oversaw the analysis and acquisition of data, were involved in the interpretation of the data, drafted the article, and critically revised the article's intellectual content. PF did the data analysis, produced the figures, was involved in the interpretation of the data, and critically revised the final manuscript. LL was involved in acquisition and interpretation of the data and critically revised the final manuscript. HD was involved in the interpretation of the data and critically revised the manuscript. AC contributed to the concept and design of the study, was involved in the interpretation of the data, and critically revised the manuscript.

n acquisition and interpretation of the data and critically revised the final manuscript. HD was involved in the interpretation of the data and critically revised the manuscript. AC contributed to the concept and design of the study, was involved in the interpretation of the data, and critically revised the manuscript. Declaration of interests CM has received grants from the UK National Institute for Health Research (NIHR), JP Moulton Charitable Foundation, and the North West Lung Research Centre Charity; and has received lecture fees from GlaxoSmithKline and Novartis. AC has received grants from the UK Medical Research Council, JP Moulton Charitable Foundation, and the North West Lung Research Centre Charity; and receives personal fees from AstraZeneca, Novartis, ThermoFisher, Regeneron/Sanofi, Bayer, ALK-Abello, GlaxoSmithKline, and Boehringer Ingelheim. AS has received grants from the UK Medical Research Council, NIHR, and European Union Framework programme 7; and has received lecture fees from GlaxoSmithKline, Chiesi, and Thermofisher Scientific. HD holds grants from Asthma UK, JP Moulton Charitable Foundation, and the North West Lung Research Centre Charity during the conduct of the study. PF and LL declare no competing interests.

Introduction Childhood obesity is considered by WHO to be one of the most serious public health challenges worldwide for the 21st century,1 and research has therefore largely focused on preventive approaches. The UK Government2 views schools as central to tackling the obesity crisis because they are an ideal setting in which to actively engage children and their families across the socioeconomic spectrum to improve diet and physical activity behaviours. In 2017, the Government has pledged to invest the revenue from the sugar levy (a tax on sugar-sweetened beverages) into school-based programmes to encourage physical activity and balanced diets.3 However, findings from systematic reviews4, 5 showed that the effectiveness of school-based obesity prevention programmes is inconclusive and contradictory: interventions are highly heterogeneous in design and most studies have methodological weaknesses, such as insufficient statistical power, high levels of attrition, differential uptake and follow-up, and only short-term follow-up (12 months on average).4, 5, 6 Research in context Evidence before this study

However, findings from systematic reviews4, 5 showed that the effectiveness of school-based obesity prevention programmes is inconclusive and contradictory: interventions are highly heterogeneous in design and most studies have methodological weaknesses, such as insufficient statistical power, high levels of attrition, differential uptake and follow-up, and only short-term follow-up (12 months on average).4, 5, 6 Research in context Evidence before this study Before the Healthy Lifestyles Programme (HeLP) was designed, a 2005 Cochrane systematic review recommended that interventions to prevent childhood obesity should have a rigorous assessment design that enables sufficiently powered analysis of what is working or not and for whom the intervention is working, and that stakeholders should be included in the development of the programme. A 2009 Cochrane systematic review of school-based obesity prevention interventions was unable to make definitive conclusions because studies were heterogeneous and only five of 38 studies followed up participants for more than 12 months. In 2011, a meta-analysis of 27 studies aiming to prevent obesity in children aged 6–12 years found some evidence for the effectiveness of combined diet and physical activity interventions; policies and strategies that appeared to be promising included providing support for teachers to implement health promotion strategies and activities in schools, and parental support that encourages healthy behaviour in children. In 2015, a review of childhood obesity prevention studies showed a moderate strength of evidence to support the effectiveness of school-based interventions. The Active for Life Year 5 (AfLY5) cluster randomised controlled trial tested a school-based intervention for children aged 9–10 years. The programme included lessons and child–parent interactive homework plans and was adapted from the American Planet Health Programme. No effect on weight status or on objectively measured physical activity or diet was found at 12 months.

controlled trial tested a school-based intervention for children aged 9–10 years. The programme included lessons and child–parent interactive homework plans and was adapted from the American Planet Health Programme. No effect on weight status or on objectively measured physical activity or diet was found at 12 months. Added value of this study HeLP was developed using an intervention mapping approach involving relevant behaviour change theories, best available evidence, and extensive involvement of teachers, head teachers, families, and children. To our knowledge, our study is the most comprehensive obesity prevention trial to date, involving a large, nationally representative sample of children aged 9–10 years and using prespecified standard methods for randomisation and analysis. The HeLP intervention was delivered with a high degree of fidelity and engaged more than 90% of children and 75% of their families. The evidence from this study therefore has internal validity and is potentially widely applicable. Implications of all the available evidence

HeLP was developed using an intervention mapping approach involving relevant behaviour change theories, best available evidence, and extensive involvement of teachers, head teachers, families, and children. To our knowledge, our study is the most comprehensive obesity prevention trial to date, involving a large, nationally representative sample of children aged 9–10 years and using prespecified standard methods for randomisation and analysis. The HeLP intervention was delivered with a high degree of fidelity and engaged more than 90% of children and 75% of their families. The evidence from this study therefore has internal validity and is potentially widely applicable. Implications of all the available evidence Our results highlight the tension facing childhood obesity prevention programmes, because schools are an ideal setting in which to deliver population-based interventions. However, taking into account the inconclusive evidence from the most recent systematic reviews and the results from both ours and the AfLY5 trial, we believe that individually focused school-based interventions targeting a single age group are unlikely to be sufficiently intense or family focused to affect the weight status of children. Future research should focus on more upstream determinants of obesity and use whole-systems approaches.

urs and the AfLY5 trial, we believe that individually focused school-based interventions targeting a single age group are unlikely to be sufficiently intense or family focused to affect the weight status of children. Future research should focus on more upstream determinants of obesity and use whole-systems approaches. In line with WHO's Health Promoting Schools framework,7 we developed the Healthy Lifestyles Programme (HeLP), consisting of activities that were compatible with the English national school curriculum and promoting messages in a manner that could affect both the wider school culture and specific behaviours of children and their families. The objective of this trial was to ascertain whether HeLP was effective in preventing childhood obesity. Methods Study design This two-arm, pragmatic, school-based, cluster randomised controlled trial with masked outcome assessment was done in 32 schools in the southwest of England. Ethics approval was given by the Peninsula College of Medicine and Dentistry Research Ethics Committee (reference number 12/03/140), and research governance approval by the Royal Devon and Exeter National Health Service Trust (study number 1304762). The full trial protocol has been published8 and is also available online.

ics approval was given by the Peninsula College of Medicine and Dentistry Research Ethics Committee (reference number 12/03/140), and research governance approval by the Royal Devon and Exeter National Health Service Trust (study number 1304762). The full trial protocol has been published8 and is also available online. Participants All state-run primary and junior schools in Devon and Plymouth (UK) with enough pupils for at least one year-5 class (children aged 9–10 years) were eligible. Schools for children whose additional needs cannot be met in a mainstream setting were excluded because they were unlikely to be teaching the standard national curriculum, around which the intervention had been designed. Schools willing to participate and fulfilling the inclusion criteria were then purposefully sampled by JL and KW to represent a range of school sizes (one to three year-5 classes), locations (urban and rural), and socioeconomic status (<19% and ≥19% of children eligible for free school meals). We aimed to have half of the schools in the trial with at least the national average proportion of pupils eligible for free schools meals (19% at the time of recruitment of schools). Before randomisation, head teachers from all schools gave written informed consent.

and ≥19% of children eligible for free school meals). We aimed to have half of the schools in the trial with at least the national average proportion of pupils eligible for free schools meals (19% at the time of recruitment of schools). Before randomisation, head teachers from all schools gave written informed consent. To accommodate the logistics and personnel required for delivering the week-long drama component of the intervention to each year-5 class, the trial ran across two cohorts (cohort 1 commenced the trial in September, 2012, and cohort 2 in September, 2013). Schools that were eligible but not sampled for the trial were asked if they were prepared to go on a waiting list, in case any of the schools allocated to participate in cohort 2 dropped out during the interim 1-year period before commencing participation. All children in all year-5 classes within each recruited school were invited to participate, and their parents or carers could choose to opt their child out before baseline measurements were taken (full details in protocol).8 All children who were on the registration list at one of the recruited schools at the start of the autumn term 2012 (for cohort 1) or 2013 (for cohort 2), and whose parents or carers did not complete an opt-out form, were classed as participants.

out before baseline measurements were taken (full details in protocol).8 All children who were on the registration list at one of the recruited schools at the start of the autumn term 2012 (for cohort 1) or 2013 (for cohort 2), and whose parents or carers did not complete an opt-out form, were classed as participants. Randomisation and masking The trial manager (JL) was responsible for recruiting schools via the Devon Association of Primary School Heads and local primary school learning community meetings. Schools were randomly allocated (1:1) to the intervention or control group using a computer-generated sequence using two school-level stratification factors: one versus more than one year-5 class and the proportion of children eligible for free school meals (<19% vs ≥19%). Randomisation was done by a statistician (RST) in the UK Clinical Research Collaboration-registered Peninsula Clinical Trials Unit immediately after all schools had been recruited in 2012, but each school's allocated group (intervention or control) was not communicated to the schools, parents, or researchers until after baseline measurements had been taken. RST ensured that numbers of control and intervention schools were equal in both cohorts to facilitate trial delivery.

ols had been recruited in 2012, but each school's allocated group (intervention or control) was not communicated to the schools, parents, or researchers until after baseline measurements had been taken. RST ensured that numbers of control and intervention schools were equal in both cohorts to facilitate trial delivery. Because of the nature of the intervention, school staff, children, and individuals delivering the intervention could not be masked to group allocation. Anthropometric measures at 18 and 24 months were collected by independent, masked, trained assessors who were not involved in the trial. We made a timeline cluster diagram for the trial to show the masking procedures for each measure at each timepoint (appendix). At the 24-month primary endpoint, when secondary schools contained a mixture of children from intervention and control primary schools, an assessment of the fidelity of assessor masking was made to ascertain whether a child had revealed their group allocation during the measurement process. If the child had revealed their group allocation in any way then this was recorded by the assessor.

xture of children from intervention and control primary schools, an assessment of the fidelity of assessor masking was made to ascertain whether a child had revealed their group allocation during the measurement process. If the child had revealed their group allocation in any way then this was recorded by the assessor. Procedures In schools assigned to the intervention group, HeLP was delivered to year-5 children (ages 9–10 years) over three school terms (roughly 12 weeks per term). HeLP consisted of four phases, which were ordered to enable and support behaviour change by targeting school and family environments and giving children the strategies and motivation to improve their snacking and activity-related behaviours (panel). Findings from the first pilot of the intervention9 (delivered to children aged 8–11 years, school years 4–6) showed that year-5 children were most receptive to the healthy lifestyle messages and engaged their parents to the greatest extent. Also, the school could more feasibly run the HeLP activities in year 5 than year 6, when the curriculum focused on standard assessment tasks. As a result, the intervention was targeted at students in year 5, while also trying to affect the wider school environment.9 The programme delivered a general healthy lifestyle message with a focus on behaviours such as the consumption of sugar-sweetened beverages, healthy and unhealthy snacking, physical activity, and reducing screen time. An overarching message promoted was the 80/20 rule, which recommended eating healthily and being active at least 80% of the time. HeLP was designed to fit in with the national curriculum at key stage 2, and all lessons and drama sessions included learning objectives relating to personal social and health education, science, numeracy, and literacy (see further details in the appendix). The development, content, and theoretical underpinning of the intervention have been published previously.10Panel Summary of intervention phases Phase 1: creating a supportive context (spring term of year 5)

to personal social and health education, science, numeracy, and literacy (see further details in the appendix). The development, content, and theoretical underpinning of the intervention have been published previously.10Panel Summary of intervention phases Phase 1: creating a supportive context (spring term of year 5) The aim is to establish relationships between all stakeholders (ie, head teachers, teachers, support staff, children, and parents) and raise awareness of the programme. Professional sports people and dancers run practical workshops and introduce the importance of healthy lifestyles to create a buzz in the school and set a positive atmosphere for future activities. At the end of this phase, children showcase the skills they have learnt in a parent assembly, in which parents are given further information about the programme by the Healthy Lifestyles Programme (HeLP) coordinator. Phase 2: intensive healthy lifestyle week (summer term of year 5)

The aim is to establish relationships between all stakeholders (ie, head teachers, teachers, support staff, children, and parents) and raise awareness of the programme. Professional sports people and dancers run practical workshops and introduce the importance of healthy lifestyles to create a buzz in the school and set a positive atmosphere for future activities. At the end of this phase, children showcase the skills they have learnt in a parent assembly, in which parents are given further information about the programme by the Healthy Lifestyles Programme (HeLP) coordinator. Phase 2: intensive healthy lifestyle week (summer term of year 5) Education lessons are delivered by the class teacher each morning and interactive drama activities by a local drama group every afternoon during the week. Short and simple homework tasks are given at the end of each session for the children to complete in time for the next session. The drama framework is built around four characters (Disorganised Duncan, Football Freddie, Snacky Sam, and Active Amy), each represented by an actor, whose attributes relate to the three key programme behaviours (reducing unhealthy snacking, increasing physical activity, and reducing sedentary activities). Children are asked to choose the character they felt they most resemble and, throughout the week, work closely with that actor to help the character change their behaviour. These sessions are dynamic and interactive and involve role play, games, dance, problem solving, food tasting, and forum theatre. The themes for each lesson and drama session are as follows: energy in and out, overcoming temptation, decision making and responsibility, food marketing, and goal setting.

their behaviour. These sessions are dynamic and interactive and involve role play, games, dance, problem solving, food tasting, and forum theatre. The themes for each lesson and drama session are as follows: energy in and out, overcoming temptation, decision making and responsibility, food marketing, and goal setting. Phase 3: personal goal setting with parental support (summer term of year 5) Children are encouraged to reflect on their own behaviours and set goals (based on the HeLP messages) with their parents. After reflection with parents, each child has a 10-min one-to-one goal-setting discussion with a HeLP coordinator. A sheet with each child's goals and the name and attributes of the character the child worked with is sent directly home to parents and a copy is also kept at school in the children's healthy lifestyles folder. Phase 4: reinforcement activities (autumn term of year 6) A range of components were used to refocus the children and their parents on the HeLP messages and behaviour change strategies. This phase includes a further lesson led by the class teacher, a drama workshop delivered by the actors, an assembly delivered by the class to the whole school about the programme, and a second one-to-one goal discussion with a HeLP coordinator.

en and their parents on the HeLP messages and behaviour change strategies. This phase includes a further lesson led by the class teacher, a drama workshop delivered by the actors, an assembly delivered by the class to the whole school about the programme, and a second one-to-one goal discussion with a HeLP coordinator. One year-5 class per school had their physical activity levels assessed using accelerometers. If a school had more than one year-5 class, a computer-generated sequence was used to randomly select one class. Children were asked to wear a waterproof triaxial accelerometer continuously (including at night) for 8 consecutive days on the wrist of their non-dominant arm. Schools assigned to the control group continued standard education provision throughout their participation in the trial, and had no access to any of the HeLP resources and scripts, which have not been published and were not made available by the research team beyond the intervention schools. Control schools each received £1000 for their participation following the collection of 18-month data. Baseline assessments were done in the autumn term of school year 5 between October and November (2012 for cohort 1 and 2013 for cohort 2). Delivery of the intervention began the following term (January, 2013, for cohort 1 and January, 2014, for cohort 2). Follow-up outcome measures were taken at 18 and 24 months after baseline. All measurements were taken at school during the school day. Fidelity of intervention delivery was assessed in relation to both content and the quality of delivery (appendix).

ng term (January, 2013, for cohort 1 and January, 2014, for cohort 2). Follow-up outcome measures were taken at 18 and 24 months after baseline. All measurements were taken at school during the school day. Fidelity of intervention delivery was assessed in relation to both content and the quality of delivery (appendix). Outcomes The primary outcome was the change in body-mass index (BMI) standard deviation score (SDS) between baseline and 24-month follow-up. BMI was calculated and converted to centiles using the LMS method for constructing normalised growth standards.11 Categorisations of underweight, normal, overweight, or obese were based on the definitions from Cole and colleagues.12 Secondary outcomes were BMI SDS at 18 months; the percentage of children classified as underweight, healthy weight, overweight, and obese at 18 and 24 months; waist circumference SDS at 18 and 24 months; percentage body fat SDS at 18 and 24 months; physical activity measured using accelerometry at 18 months; and self-reported scores for the number of different types of energy-dense snacks, healthy snacks, healthy foods (positive food markers), and unhealthy foods (negative food markers) consumed per day using the validated Food Intake Questionnaire (FIQ)13 at 18 months (appendix). Details of methods of data collection for the anthropometric and behavioural measures are in the appendix. An adverse event was considered to include unusual dieting or physical activity behaviours or noticeable weight loss. Any adverse event could be reported by school staff, parents, HeLP coordinators, or actors.

Secondary outcomes were BMI SDS at 18 months; the percentage of children classified as underweight, healthy weight, overweight, and obese at 18 and 24 months; waist circumference SDS at 18 and 24 months; percentage body fat SDS at 18 and 24 months; physical activity measured using accelerometry at 18 months; and self-reported scores for the number of different types of energy-dense snacks, healthy snacks, healthy foods (positive food markers), and unhealthy foods (negative food markers) consumed per day using the validated Food Intake Questionnaire (FIQ)13 at 18 months (appendix). Details of methods of data collection for the anthropometric and behavioural measures are in the appendix. An adverse event was considered to include unusual dieting or physical activity behaviours or noticeable weight loss. Any adverse event could be reported by school staff, parents, HeLP coordinators, or actors. Statistical analysis Our sample size calculation assumed a mean of 35 children aged 9–10 years per school, with coefficient of variance of 0·5 and an intraclass correlation coefficient of 0·02. To have 90% power, with a two-sided 5% significance level, to detect a between-group difference in BMI SDS of 0·25 units at 24 months, assuming an SD of 1·3 and adjusting for baseline BMI SDS (assumed within-child correlation of 0·8), we needed to have 24-month outcome data from at least 762 children. Allowing for up to 20% loss to follow-up, we aimed to recruit 28 schools with at least 952 children.

p difference in BMI SDS of 0·25 units at 24 months, assuming an SD of 1·3 and adjusting for baseline BMI SDS (assumed within-child correlation of 0·8), we needed to have 24-month outcome data from at least 762 children. Allowing for up to 20% loss to follow-up, we aimed to recruit 28 schools with at least 952 children. The primary analyses were done in children with BMI data available for both baseline the 24-month follow-up by a statistician masked to the allocated group. Because of the high levels of completeness of data and low proportion of children categorised as non-compliers, the multiple imputation approach for handling missing outcome data was replaced by a best-case and worst-case scenario, and the planned complier average causal effect analyses were dropped.14 All comparative analyses allowed for the clustering of children within schools15 using a likelihood-based random-effects regression modelling approach that uses all available data and provides valid estimates of the effect of the intervention, when data are assumed to be missing at random. Most of the outcomes were of a continuous nature and thus linear models were fitted. Weight status was analysed using a random-effects ordinal logistic regression model with three categories (underweight or healthy weight, overweight, and obese) and a random-effects logistic regression model with two categories (underweight or healthy weight and overweight or obese); for simplicity, only the results of the logistic regression models are reported here.

ordinal logistic regression model with three categories (underweight or healthy weight, overweight, and obese) and a random-effects logistic regression model with two categories (underweight or healthy weight and overweight or obese); for simplicity, only the results of the logistic regression models are reported here. All primary comparative analyses were adjusted for the two school-level stratification factors (proportion of children eligible for free school meals and number of year-5 classes), cohort, sex, and the baseline values for the outcome under consideration, when available. Adjusted between-group mean differences (intervention minus control) and odds ratios (intervention vs control), with corresponding 95% CIs, were calculated for all outcomes. p values are two sided and were considered significant at 0·05 or less. Between-group differences with adjustment only for clustering are given for completeness.15 The intraclass correlation coefficient (with 95% CI) from the random-effects regression models are reported for all outcomes.

e calculated for all outcomes. p values are two sided and were considered significant at 0·05 or less. Between-group differences with adjustment only for clustering are given for completeness.15 The intraclass correlation coefficient (with 95% CI) from the random-effects regression models are reported for all outcomes. Additional preplanned exploratory subgroup analyses were done to assess whether any effect of the HeLP intervention on the primary outcome was modified by sex, baseline BMI SDS, number of year-5 classes within a school, child-level socioeconomic status, or trial entry time (ie, cohort effect). We also fitted a repeated measures model to all the BMI SDS data at baseline, 18 months, and 24 months, including effects of time and the interaction term between allocated group and time, to assess whether the effect of the intervention differed over time. For the physical activity analysis, children were included if they had data on at least three weekdays and one weekend day, each with a minimum of 10 h per day.16 In the analyses, non-wear of the accelerometer was established by at least two accelerometer axes with an SD less than 13 mg and a range less than 50 mg over a 60-min period, using moving increments of 15 mins.17 Accelerometers were set to record at 85·7 Hz and data were downloaded using GeneActiv PC software, version 1.4, and analysed using the GGIR software package for R.

lished by at least two accelerometer axes with an SD less than 13 mg and a range less than 50 mg over a 60-min period, using moving increments of 15 mins.17 Accelerometers were set to record at 85·7 Hz and data were downloaded using GeneActiv PC software, version 1.4, and analysed using the GGIR software package for R. We also did parallel economic and process analyses (see analysis plan in the protocol), which will be reported separately. A detailed statistical analysis plan has been published.14 All analyses were done in Stata, version 14.0, unless otherwise stated. This trial is registered with International Standard Randomised Controlled Trials, number ISRCTN15811706. Role of the funding source The funders had no role in study design, data collection, data analysis, data interpretation, or writing of the report. JL, KW, and SC had full access to all the data. All authors commented on drafts and approved the final report, and JL had final responsibility for the decision to submit for publication.

unding source The funders had no role in study design, data collection, data analysis, data interpretation, or writing of the report. JL, KW, and SC had full access to all the data. All authors commented on drafts and approved the final report, and JL had final responsibility for the decision to submit for publication. Results Between March 21, 2012, and Sept 30, 2013, 36 eligible schools were identified, of which four were placed on the waiting list. We recruited 32 schools with 1371 eligible children, of whom 1324 participated in the study (figure 1). 16 schools (676 children) were assigned to the intervention group and 16 schools (648 children) to the control group. We compared characteristics of the primary schools in the HeLP trial with other primary schools in Devon and England (appendix p 2). HeLP schools had a similar average number of pupils, deprivation, and academic achievement to English schools; however, the proportion of pupils with English as a second language was significantly lower than the national average (4·1% in HeLP schools vs 16·8% in all schools in England), although it was nearly double the proportion in Devon schools as a whole, which is 2·6%. The intervention and control groups had similar school-level and child-level baseline characteristics, including physical activity and food intake questionnaire scores (table 1). At baseline, although anthropometric measurements between the groups were largely similar, a greater proportion of children in the intervention group were overweight or obese than in the control group (table 1).Figure 1 Trial profile

teristics, including physical activity and food intake questionnaire scores (table 1). At baseline, although anthropometric measurements between the groups were largely similar, a greater proportion of children in the intervention group were overweight or obese than in the control group (table 1).Figure 1 Trial profile *Two schools that had been allocated to cohort 2 withdrew while waiting to commence the trial and so were replaced with two of the four schools on the waiting list before cohort 2 commenced the trial. All schools that started the trial remained in the trial and so all the randomised clusters are present at baseline and at each follow-up point. BMI=body-mass index. Table 1 Baseline characteristics of participating schools and children

*Two schools that had been allocated to cohort 2 withdrew while waiting to commence the trial and so were replaced with two of the four schools on the waiting list before cohort 2 commenced the trial. All schools that started the trial remained in the trial and so all the randomised clusters are present at baseline and at each follow-up point. BMI=body-mass index. Table 1 Baseline characteristics of participating schools and children Intervention group Control group Cluster level Number of schools 16 16 Number of participating children per school 35·4 (26·5–50·0) 33·5 (28·5–51·0) School IMD* 14 380 (12 806–21 446) 13 341 (9208–21 785) Number of year-5 classes Single class 8 (50%) 9 (56%) More than one class 8 (50%) 7 (44%) Free school meals <19% of pupils 9 (56%) 9 (56%) ≥19% of pupils 7 (44%) 7 (44%) Cohort Cohort 1 8 (50%) 8 (50%) Cohort 2 8 (50%) 8 (50%) Individual level Number of children 676 648 Age, years 9·8 (0·3) 9·7 (0·3) Sex Female 336 (50%) 343 (53%) Male 340 (50%) 305 (47%) Child IMD* 16 060 (12347–21957) 13 171 (6741–20 882) BMI SDS 0·32 (1·16) 0·18 (1·14) Waist circumference SDS 0·72 (1·11) 0·55 (1·15) Percentage body fat SDS −0·61 (2·18) −0·63 (2·38) Percentage body fat SDS (excluding extreme body fat)† −0·39 (1·62) −0·46 (1·52) Weight status‡ Underweight 11 (2%) 10 (2%) Healthy 479 (72%) 483 (75%) Overweight 81 (12%) 69 (11%) Obese 98 (15%) 81 (13%) Missing data 7 (1%) 5 (1%) Physical activity§ Weekly acceleration, mg 49·0 (11·3) 49·6 (10·9) Daily total, min 182·7 (36·7) 185·0 (34·7) Daily light, min 129·4 (24·7) 131·1 (24·2) Daily moderate, min 40·0 (12·1) 40·4 (11·4) Daily moderate to vigorous, min 53·3 (16·8) 53·9 (16·2) Daily vigorous, min 13·3 (6·2) 13·5 (6·2) Daily sedentary, min 780·4 (36·1) 778·2 (34·0) Food intake questionnaire scores, all days of the week Daily energy-dense snacks 4·2 (2·2) 4·1 (2·2) Daily healthy snacks 3·3 (1·6) 3·1 (1·6) Daily positive food markers 6·0 (2·7) 5·7 (2·5) Daily negative food markers 6·8 (3·4) 6·8 (3·3) Food intake questionnaire scores, weekdays Daily energy-dense snacks 4·0 (2·4) 4·0 (2·4) Daily healthy snacks 3·4 (1·8) 3·2 (1·7) Daily positive food markers 6·1 (2·9) 5·7 (2·8) Daily negative food markers 6·5 (3·7) 6·7 (3·8) Food intake questionnaire scores, weekend days Daily energy-dense snacks 4·6 (2·5) 4·4 (2·4) Daily healthy snacks 3·2 (1·9) 2·9 (1·8) Daily positive food markers 6·0 (3·1) 5·5 (2·9) Daily negative food markers 7·7 (4·0) 7·1 (3·6) Data are n (%), mean (SD), or median (IQR). IMD=index of multiple deprivation. BMI=body-mass index. SDS=standard deviation score.

nd days Daily energy-dense snacks 4·6 (2·5) 4·4 (2·4) Daily healthy snacks 3·2 (1·9) 2·9 (1·8) Daily positive food markers 6·0 (3·1) 5·5 (2·9) Daily negative food markers 7·7 (4·0) 7·1 (3·6) Data are n (%), mean (SD), or median (IQR). IMD=index of multiple deprivation. BMI=body-mass index. SDS=standard deviation score. * School IMD is related to the school's postcode and child IMD is related to child's home postcode.18 † After excluding extreme body fat absolute SDS ≥5. ‡ At baseline, height and weight measurements were available for 669 (99%) of 676 children in the intervention and 643 (99%) of 648 in the control group. Weight status categories defined using the Public Health England definitions12 (underweight ≤2nd UK National BMI percentile relevant to the UK 1990 reference data, healthy >2nd and <85th BMI percentile, overweight ≥85th and <95th BMI percentile, and obese ≥95th BMI percentile). § n=428 in intervention group and n=458 in control group.

‡ At baseline, height and weight measurements were available for 669 (99%) of 676 children in the intervention and 643 (99%) of 648 in the control group. Weight status categories defined using the Public Health England definitions12 (underweight ≤2nd UK National BMI percentile relevant to the UK 1990 reference data, healthy >2nd and <85th BMI percentile, overweight ≥85th and <95th BMI percentile, and obese ≥95th BMI percentile). § n=428 in intervention group and n=458 in control group. All 32 schools completed the trial. All schools in the intervention group completed or nearly completed the whole programme and the quality of delivery in all schools was at or above the established appropriate level (appendix). 629 (93%) of the 676 children in the intervention group were categorised as compliers (ie, they received at least four of the five drama sessions and the one-to-one goal-setting discussion in phase 3). No notable differences in uptake were seen between the two cohorts (appendix). 353 (52%) of the 676 children had family attending at least one parent event and 652 (96%) children set goals with the HeLP coordinator in phase 3. 411 (63%) of these 652 children had parental support, shown by a parent signature on the goal-setting sheet or written comments about how the parent would support the child in achieving their goals.

ren had family attending at least one parent event and 652 (96%) children set goals with the HeLP coordinator in phase 3. 411 (63%) of these 652 children had parental support, shown by a parent signature on the goal-setting sheet or written comments about how the parent would support the child in achieving their goals. 1244 children were included in the primary analysis of BMI SDS because both baseline and 24-month BMI data were available for them (figure 1). In the measurement training sessions before anthropometric measures were taken, inter-rater reliability for height and waist circumference was high (coefficients of variations were, respectively, 0·2% and 1·3% at baseline, 0·1% and 1·2% at 18 months, and 0·1% and 0·4% at 24 months). No child had reported his or her allocated group to the masked assessor at 24-month follow-up. Of the 886 children who wore accelerometers, 851 (96%) had usable physical activity data files (ie, files could be downloaded and were not corrupted) at baseline and 788 (89%) had usable data at 18 months; similarly, the number of children with valid physical activity data after the application of the minimum wear requirements (three weekdays and one weekend day with a minimum of 10 h of wear time per day) was 830 (94%) at baseline and 745 (84%) at 18 months. 701 (79%) of the 886 children achieved the full 7 days of 10 h wear time per day. No evidence suggested differences between the groups in terms of the completeness of outcome measures throughout the trial, although more children in the intervention group were lost to follow-up than in the control group (figure 1). No differences were noted between control and intervention schools at either baseline or 18 months in terms of the number and type of school nutrition and physical activity policies they had in place (data not shown).

h more children in the intervention group were lost to follow-up than in the control group (figure 1). No differences were noted between control and intervention schools at either baseline or 18 months in terms of the number and type of school nutrition and physical activity policies they had in place (data not shown). Mean BMI SDS at 24 months was 0·35 (SD 1·25) in children in the intervention group and 0·22 (1·22) in those in the control group (table 2). With adjustment for school-level clustering, baseline BMI scores, sex, cohort, and number of year-5 classes and socioeconomic status of each school, the mean difference in BMI SDS score (intervention–control) at 24 months was −0·02 (95% CI −0·09 to 0·05, p=0.57).Table 2 Primary and secondary anthropometric outcomes at 18 and 24 months

for school-level clustering, baseline BMI scores, sex, cohort, and number of year-5 classes and socioeconomic status of each school, the mean difference in BMI SDS score (intervention–control) at 24 months was −0·02 (95% CI −0·09 to 0·05, p=0.57).Table 2 Primary and secondary anthropometric outcomes at 18 and 24 months Intervention group Control group Mean difference (intervention–control) or odds ratio (95% CI) p value* N† Mean (SD) or (%) N† Mean (SD) or (%) Adjusted for clustering only Fully adjusted‡ 18 months BMI SDS 644 0·32 (1·23) 621 0·20 (1·23) 0·11 (−0·12 to 0·33) −0·02 (−0·08 to 0·05) 0·61 Waist circumference SDS 645†† 0·69 (1·18) 620§ 0·57 (1·15) 0·08 (−0·15 to 0·32) −0·07 (−0·27 to 0·12) 0·44 Percentage body fat SDS All children 644 −0·99 (2·23) 619§ −0·98 (2·03) −0·02 (−0·38 to 0·35) −0·02 (−0·25 to 0·22) 0·90 After exclusion of extreme values¶ 618 −0·74 (1·84) 593 −0·75 (1·73) 0·01 (−0·29 to 0·31) −0·02 (−0·16 to 0·12) 0·77 Weight status|| Underweight and healthy weight 458 71% 463 75% NA NA NA Overweight 87 14% 78 13% NA NA NA Obese 99 15% 80 13% NA NA NA Overweight and obese 186 29% 158 25% 1·18** (0·80 to 1·72) 1·05** (0·58 to 1·88) 0·88 24 months BMI SDS 630 0·35 (1·25) 620 0·22 (1·22) 0·11 (−0·11 to 0·33) −0·02 (−0·09 to 0·05) 0·57 Waist circumference SDS 629§ 0·63 (1·24) 618§ 0·54 (1·21) 0·09 (−0·15 to 0·33) −0·05 (−0·23 to 0·13) 0·56 Percentage body fat SDS All children 629§ −0·78 (2·16) 620 −0·78 (1·89) −0·02 (−0·37 to 0·33) −0·04 (−0·29 to 0·22) 0·76 After exclusion of extreme values¶ 612 −0·59 (1·84) 607 −0·65 (1·69) 0·04 (−0·23 to 0·32) −0·02 (−0·17 to 0·13) 0·79 Weight status|| Underweight and healthy weight 436 69% 455 73% NA NA NA Overweight 89 14% 84 14% NA NA NA Obese 105 17% 81 13% NA NA NA Overweight and obese 194 31% 165 27% 1·19** (0·82 to 1·71) 1·09** (0·70 to 1·69) 0·72 BMI=body-mass index. SDS=standard deviation score. NA=not applicable; the logistic regression to produce odd ratios was only applicable to the combined overweight and obese category, dichotomised into levels of overweight and obese versus normal and underweight.

165 27% 1·19** (0·82 to 1·71) 1·09** (0·70 to 1·69) 0·72 BMI=body-mass index. SDS=standard deviation score. NA=not applicable; the logistic regression to produce odd ratios was only applicable to the combined overweight and obese category, dichotomised into levels of overweight and obese versus normal and underweight. * Fully adjusted mean difference. † N is the total number of children from whom we collected data at that timepoint. ‡ Estimated using random-effects linear or logistic regression models (comparing overweight or obese with underweight or healthy weight) to account for clustering among children within the same school, with adjustment for stratification variables (number of year-5 classes and proportion of children eligible for free school meals), cohort, sex, and baseline measure of outcome under consideration. § Some data for some children were not collected because they were absent on days of assessment or they left or moved between schools. ¶ After excluding extreme body fat absolute SD values ≥5. || Weight status categories defined using the Public Health England definitions12 (underweight ≤2nd UK National BMI percentile relevant to the UK 1990 reference data, healthy >2nd and <85th BMI percentile, overweight ≥85th and <95th BMI percentile, and obese ≥95th BMI percentile). ** Results from logistic regression analysis. †† At 18 months, one child had waist circumference measurement but no weight measurement, so BMI could not be calculated.

|| Weight status categories defined using the Public Health England definitions12 (underweight ≤2nd UK National BMI percentile relevant to the UK 1990 reference data, healthy >2nd and <85th BMI percentile, overweight ≥85th and <95th BMI percentile, and obese ≥95th BMI percentile). ** Results from logistic regression analysis. †† At 18 months, one child had waist circumference measurement but no weight measurement, so BMI could not be calculated. According to the repeated measures model, no significant difference in mean BMI SDS existed between the two allocated groups at baseline (0·30 [95% CI 0·18 to 0·41] in the intervention group and 0·18 [0·06 to 0·30] in the control group, p=0·17; figure 2). The BMI SDS predicted by the model was 0·30 (95% CI 0·18 to 0·41) in the intervention group and 0·21 (0·09 to 0·33) in the control group at 18 months, increasing to 0·33 (0·21 to 0·45) in the intervention group and 0·23 (95% CI 0·11 to 0·35) in the control group by 24 months (figure 2). The sensitivity analyses to explore assumptions about missing primary outcome data produced results that were consistent with the primary analysis (appendix). We found no evidence that the intervention effect was modified in any of the prespecified subgroups (appendix).Figure 2 Predicted marginal BMI SDS with 95% CIs in the two groups across timepoints

e assumptions about missing primary outcome data produced results that were consistent with the primary analysis (appendix). We found no evidence that the intervention effect was modified in any of the prespecified subgroups (appendix).Figure 2 Predicted marginal BMI SDS with 95% CIs in the two groups across timepoints Data are derived from the repeated measures, allowing for hierarchical clustering by child within each school, modelling the within-child covariance between fixed timepoints as an autoregressive pattern of order one. BMI=body-mass index. SDS=standard deviation score.

e assumptions about missing primary outcome data produced results that were consistent with the primary analysis (appendix). We found no evidence that the intervention effect was modified in any of the prespecified subgroups (appendix).Figure 2 Predicted marginal BMI SDS with 95% CIs in the two groups across timepoints Data are derived from the repeated measures, allowing for hierarchical clustering by child within each school, modelling the within-child covariance between fixed timepoints as an autoregressive pattern of order one. BMI=body-mass index. SDS=standard deviation score. No significant differences were seen between the groups in any of the other anthropometric outcomes at either 18 or 24 months (table 2), nor any of the physical activity outcomes at 18 months (table 3). The adjusted means of FIQ scores (both weekly and weekday) for energy-dense snacks and negative food markers were lower in the intervention group than in the control group (table 4). The discrete values of the weekday and weekend scores, bounded by zero, suggested that the assumptions for modelling in a linear model might not be fully met, despite the apparent symmetrical, normal shape of the data in the plots. Therefore, random-effects ordinal logistic regression models were also fitted to these outcomes. A positive effect of the intervention was still seen on the weekday scores for energy-dense snacks and negative food markers (data not shown); however, the p values are close to 0·05 and the difference could be due to chance.Table 3 Primary intention-to-treat analyses of secondary physical activity outcome measures assessed at 18 months after baseline

tion was still seen on the weekday scores for energy-dense snacks and negative food markers (data not shown); however, the p values are close to 0·05 and the difference could be due to chance.Table 3 Primary intention-to-treat analyses of secondary physical activity outcome measures assessed at 18 months after baseline Intervention group (n=359) Control group (n=386) Mean difference (intervention–control) (95% CI) p value† Adjusted for clustering only Fully adjusted* Weekly acceleration, mg 52·14 (13·95) 51·47 (12·95) 0·53 (−2·18 to 3·24) 0·57 (−1·58 to 2·72) 0·59 Daily total, min 199·71 (43·94) 198·05 (40·20) 1·23 (−8·24 to 10·70) 1·26 (−6·84 to 9·36) 0·75 Daily light, min 141·72 (27·80) 141·07 (27·09) 0·43 (−5·87 to 6·73) 0·70 (−4·73 to 6·13) 0·79 Daily moderate, min 44·26 (16·24) 43·46 (13·43) 0·68 (−2·41 to 3·78) 0·41 (−2·28 to 3·09) 0·76 Daily moderate to vigorous, min 57·99 (22·34) 56·98 (19·39) 0·85 (−3·24 to 4·94) 0·56 (−2·76 to 3·89) 0·73 Daily vigorous, min 13·73 (7·66) 13·52 (7·38) 0·15 (−1·33 to 1·63) 0·15 (−1·01 to 1·3) 0·80 Daily sedentary, min 764·50 (43·29) 766·36 (39·88) −1·46 (−10·91 to 8·00) −1·39 (−9·45 to 6·68) 0·73 Data are mean (SD) unless specified otherwise. * Estimated using random-effects linear regression models to account for clustering among children within the same school, with adjustment for stratification variables (number of year-5 classes and proportion of children eligible for free school meals), cohort, sex, and baseline measure of the outcome under consideration. † Fully adjusted mean difference.

* Estimated using random-effects linear regression models to account for clustering among children within the same school, with adjustment for stratification variables (number of year-5 classes and proportion of children eligible for free school meals), cohort, sex, and baseline measure of the outcome under consideration. † Fully adjusted mean difference. Table 4 Food intake questionnaire outcomes at 18 months Intervention group Control group Mean difference (intervention–control) (95% CI) p value† N Mean (SD) N Mean (SD) Adjusted for clustering only Fully adjusted*

* Estimated using random-effects linear regression models to account for clustering among children within the same school, with adjustment for stratification variables (number of year-5 classes and proportion of children eligible for free school meals), cohort, sex, and baseline measure of the outcome under consideration. † Fully adjusted mean difference. Table 4 Food intake questionnaire outcomes at 18 months Intervention group Control group Mean difference (intervention–control) (95% CI) p value† N Mean (SD) N Mean (SD) Adjusted for clustering only Fully adjusted* Weekly food intake questionnaire scores Daily energy-dense snacks 646 3·72 (1·86) 624 4·06 (2·07) −0·29 (−0·64 to 0·06) −0·37 (−0·66 to −0·07) 0·017 Daily healthy snacks 637 3·61 (1·63) 617 3·30 (1·50) 0·31 (0·02 to 0·60) 0·22 (−0·04 to 0·47) 0·092 Daily negative food markers 647 5·90 (2·73) 624 6·38 (3·00) −0·40 (−0·94 to 0·14) −0·47 (−0·91 to −0·02) 0·041 Daily positive food markers 647 6·20 (2·36) 624 5·77 (2·31) 0·42 (0·01 to 0·84) 0·26 (−0·12 to 0·64) 0·17 Weekday food intake questionnaire scores Daily energy-dense snacks 647 3·54 (2·03) 625 3·99 (2·27) −0·41 (−0·83 to 0·01) −0·47 (−0·84 to −0·11) 0·013 Daily healthy snacks 645 3·69 (1·77) 624 3·38 (1·64) 0·30 (−0·04 to 0·64) 0·23 (−0·08 to 0·54) 0·14 Daily negative food markers 647 5·54 (2·94) 625 6·21 (3·28) −0·61 (−1·25 to 0·03) −0·64 (−1·17 to −0·11) 0·020 Daily positive food markers 647 6·28 (2·55) 625 5·87 (2·52) 0·39 (−0·09 to 0·86) 0·27 (−0·18 to 0·73) 0·23 Weekend food intake questionnaire scores Daily energy-dense snacks 647 4·17 (2·21) 626 4·26 (2·35) 0·01 (−0·37 to 0·37) −0·10 (−0·43 to 0·24) 0·56 Daily healthy snacks 639 3·42 (1·83) 620 3·12 (1·73) 0·31 (0·02 to 0·59) 0·23 (−0·04 to 0·50) 0·086 Daily negative food markers 648 6·79 (3·24) 626 6·82 (3·36) 0·12 (−0·44 to 0·68) −0·07 (−0·56 to 0·42) 0·77 Daily positive food markers 648 6·00 (2·66) 626 5·52 (2·64) 0·50 (0·05 to 0·95) 0·36 (−0·13 to 0·84) 0·15 * Estimated using random-effects linear regression models to account for clustering among children within the same school, with adjustment for stratification variables (number of year-5 classes and proportion of children eligible for free school meals), cohort, sex, and baseline measure of the outcome under consideration.

stimated using random-effects linear regression models to account for clustering among children within the same school, with adjustment for stratification variables (number of year-5 classes and proportion of children eligible for free school meals), cohort, sex, and baseline measure of the outcome under consideration. † Fully adjusted mean difference. The intraclass correlation coefficient for BMI SDS at 24 months was 0·014 (95% CI 0·003–0·069; appendix p 12). Three children withdrew from the trial (two from the control group and one from the intervention group), and one adverse event was reported by a concerned parent about her child's eating and activity behaviours (overexercising and restricting food intake). After discussion with the chief investigator, the parent was happy for their child to remain in the study and continue to participate in the intervention.

ntion group), and one adverse event was reported by a concerned parent about her child's eating and activity behaviours (overexercising and restricting food intake). After discussion with the chief investigator, the parent was happy for their child to remain in the study and continue to participate in the intervention. Discussion The risk of childhood obesity is related to a complex interaction of factors at the individual, family, school, and societal levels. The HeLP intervention was developed using intervention mapping based on previous evidence of effective approaches to modifying children's risk factors for obesity and creating supportive school and home environments for healthy behaviours; it was extensively piloted to ensure acceptability and feasibility.10, 19, 20 In this large school-based cluster randomised controlled trial we showed high fidelity to intervention delivery and participation by children and families, and successfully collected data on 84–96% of children for all outcome measures. We found no evidence of an intervention effect on the primary outcome of BMI SDS at 24 months, nor on any of the objectively measured anthropometric or physical activity outcomes. Based on self-report data, there was some weak evidence of a small but significant difference in favour of the intervention group in the mean number of different types of unhealthy snacks (energy dense) and unhealthy foods (negative markers) consumed.

e objectively measured anthropometric or physical activity outcomes. Based on self-report data, there was some weak evidence of a small but significant difference in favour of the intervention group in the mean number of different types of unhealthy snacks (energy dense) and unhealthy foods (negative markers) consumed. Evidence from systematic reviews5, 21 suggests that some school-based intervention programmes that target physical activity and diet and involve activities to engage families have a modest effect on weight outcomes; however, the reviews identify significant between-study heterogeneity and acknowledge that most of the included studies have a moderate-to-high risk of bias. The most recent, methodologically rigorous, UK trial of a school-based intervention aimed at increasing physical activity, reducing sedentary behaviour, and increasing fruit and vegetable consumption in the same age group as HeLP, the Active for Life Year 5 trial,22 reported no difference between children in the intervention and control groups in the three primary outcomes (accelerometer-assessed moderate to vigorous physical activity, accelerometer-assessed sedentary activity, self-reported fruit and vegetable servings) or in the secondary outcome of weight status at 12 months.

2 reported no difference between children in the intervention and control groups in the three primary outcomes (accelerometer-assessed moderate to vigorous physical activity, accelerometer-assessed sedentary activity, self-reported fruit and vegetable servings) or in the secondary outcome of weight status at 12 months. In the HeLP trial, we aimed to address the methodological shortcomings identified in other studies and assess an intervention that included the behaviour change techniques believed most likely to affect identified causal pathways for obesity.23 It also aimed to engage children, families, and schools. We specifically sought to minimise key sources of bias, including recruitment, performance, and detection biases, by recruiting schools and children and collecting baseline measures before randomisation (to reduce differential uptake), capturing evidence of changes in school policies around food or physical activity during the trial, and using assessors who were masked to group allocation to measure the anthropometric outcomes. Although the FIQ was completed before revealing group allocation at baseline, the children were aware of their group allocation at the 18-month and 24-month follow-ups (appendix p 3). We also recognise that the HeLP coordinators collected measurements in both control and intervention schools, so contamination in the control schools might have occurred. However, the interaction between the coordinators and the control schools and children was minimal compared with that in the intervention schools, so taking the measurements probably did not affect obesity-related behaviours of the children to any great extent.

contamination in the control schools might have occurred. However, the interaction between the coordinators and the control schools and children was minimal compared with that in the intervention schools, so taking the measurements probably did not affect obesity-related behaviours of the children to any great extent. Sample size calculations, which allowed for the anticipated level of clustering as estimated from the exploratory trial and English National Child Measurement Programme (NCMP) data,19, 24 suggested that the trial needed outcome measures from 762 children at 24-month follow-up to detect a clinically meaningful difference in BMI SDS. However, the larger number of children per school than that anticipated, as well as successful trial recruitment and retention, meant that primary outcome data were available for 1250 children. Only a low risk of attrition bias existed in the study because few eligible children (34 [2%] of 1371) were opted out by their parents or carers and we achieved exceptional levels of follow-up at both 18 and 24 months for all outcome measures.

retention, meant that primary outcome data were available for 1250 children. Only a low risk of attrition bias existed in the study because few eligible children (34 [2%] of 1371) were opted out by their parents or carers and we achieved exceptional levels of follow-up at both 18 and 24 months for all outcome measures. Reviews of school-based obesity prevention and obesity management trials in children have shown low participation, differential dropout, and high loss to follow-up.5, 6, 21, 25 For example, completeness of anthropometric data in school-based obesity prevention programmes has ranged from 70% to 80% for follow-up of 24 months or more,26, 27, 28 and the percentage of children providing a representative pattern of their physical activity levels across the entire week (at least three weekdays and one weekend day of 10 h wear time) tends to be much lower (40–60%).16 In the HeLP trial, 84% of children met this minimum wear time criteria and 79% provided data on 7 days, showing one of the most complete follow-ups and best compliance with physical activity assessment of obesity prevention trials in children of this age group. We attribute this to the extensive stakeholder involvement in the intervention development, trial design and delivery, and building of trusting and supportive relationships with schools, children, and families.20, 29

ompliance with physical activity assessment of obesity prevention trials in children of this age group. We attribute this to the extensive stakeholder involvement in the intervention development, trial design and delivery, and building of trusting and supportive relationships with schools, children, and families.20, 29 We weighted school recruitment to achieve a similar proportion of pupils eligible to receive free school meals as the national average, which is higher than the average for Devon (12·7%). 14 (44%) of the 32 participating schools had more than 19% pupils eligible for free school meals. Participating schools were larger than the average primary school in Devon, but in other respects, schools were representative of Devon and the anthropometric data from the children in the trial were broadly similar to the Devon NCMP year-6 data (no county-level data are available for year-5 classes because these measurements are taken in reception and year 6 only).30 The representative sample gives us confidence that the results are likely to be applicable to a wider population. We used the proportion of children with English as a second language as a proxy of ethnicity, and, although the included schools reflect the proportion of children from minority ethnic groups typical for Devon (6%), this value is substantially lower than the average proportion in England (28%; appendix p 2).31

ulation. We used the proportion of children with English as a second language as a proxy of ethnicity, and, although the included schools reflect the proportion of children from minority ethnic groups typical for Devon (6%), this value is substantially lower than the average proportion in England (28%; appendix p 2).31 We found that a theoretically informed complex intervention, which was feasible and acceptable to schools, children, and their families and achieved a high level of engagement, failed to change diet and physical activity behaviours and had no effect on weight status. Schools are ideal locations for childhood obesity prevention programmes given their near-universal reach of children across the socioeconomic spectrum; however, the capacity of such programmes to affect family behaviour patterns is poor. Children aged 9 and 10 years spend most of their time in either the school or family environment and it was these two environments we sought to affect, giving children the necessary skills to identify and make healthy diet and activity choices and engage their parents in supporting these behaviours. We gave the children autonomy to select which behaviours they wished to change and encouraged their families to identify how they would support their child to achieve their goals. However, the programme did not affect BMI SDS or physical activity, suggesting that we were unsuccessful in our overarching aim to affect both the school and family environments. Although HeLP used several activities to directly engage parents as well as activities to engage other year groups within the school, the programme did not explicitly seek to affect school policies or physical aspects of the school environment. Furthermore, in seeking to minimise the burden of delivery for schools, the use of external delivery personnel for much of the programme might have minimised any effect on school culture. However, we think schools are unlikely to find a more intensive programme feasible or acceptable to implement.

e school environment. Furthermore, in seeking to minimise the burden of delivery for schools, the use of external delivery personnel for much of the programme might have minimised any effect on school culture. However, we think schools are unlikely to find a more intensive programme feasible or acceptable to implement. We believe that these findings, and results from other large, rigorous studies, call into question the likelihood that individually focused, school-based obesity prevention programmes can ever be sufficient to reduce the risks of obesity in primary school children. In 2015, The Lancet's second Obesity Series called for an “urgent rethinking of the causes, enablers, and barriers to change” by focusing on the “reciprocal nature of the interaction between the environment and the individual”,32 in which feedback loops perpetuate food choices and behaviours. Schools have an important role in creating supportive social and physical environments; however, unless upstream determinants of obesity are also addressed, families are unlikely to feel supported or motivated to change their behaviours. Such whole-systems approaches to childhood obesity prevention are theoretically attractive, but both their practical application and evidence for their effectiveness are currently absent and will require rigorous investigation. Supplementary Material Supplementary appendix

We believe that these findings, and results from other large, rigorous studies, call into question the likelihood that individually focused, school-based obesity prevention programmes can ever be sufficient to reduce the risks of obesity in primary school children. In 2015, The Lancet's second Obesity Series called for an “urgent rethinking of the causes, enablers, and barriers to change” by focusing on the “reciprocal nature of the interaction between the environment and the individual”,32 in which feedback loops perpetuate food choices and behaviours. Schools have an important role in creating supportive social and physical environments; however, unless upstream determinants of obesity are also addressed, families are unlikely to feel supported or motivated to change their behaviours. Such whole-systems approaches to childhood obesity prevention are theoretically attractive, but both their practical application and evidence for their effectiveness are currently absent and will require rigorous investigation. Supplementary Material Supplementary appendix Acknowledgments We thank all the pupils, teachers, head teachers, teaching assistants, and parents and carers for participating in the study. We thank the HeLP coordinators, drama teams, director of Headbangers Theatre Company, Just4Funk, Attik Dance, Exeter Chiefs, and Plymouth Raiders for their enthusiasm and support in delivering the programme. We thank all the HeLP study staff, trained assessors, and trainers who delivered assessment training and data management, and administrative staff who provided support throughout the trial. We are indebted to our Public Advisory Group, whose advice and ongoing support were invaluable in ensuring the trial design and intervention was feasible and acceptable to schools, children, and their families. We also thank the Chair and Trial Steering Committee members for their advice and support throughout the trial. We would also like to acknowledge the UK National Institute for Health Research Collaboration for Leadership in Applied Health, Research and Care, South West Peninsula, for providing non-financial methodological support during the transition from the exploratory trial to the definitive analysis.

t throughout the trial. We would also like to acknowledge the UK National Institute for Health Research Collaboration for Leadership in Applied Health, Research and Care, South West Peninsula, for providing non-financial methodological support during the transition from the exploratory trial to the definitive analysis. Contributors KW was the chief investigator. JL, SC, SL, CG, SGD, RST, and KW were involved in all stages of the HeLP trial, including conception, design, interpretation of data, and in seeking funding. VP, ER, and RT provided stakeholder input into the development phases and the trial. JL and KW were responsible for the conduct of the study. JL led the development of the HeLP intervention, managed the delivery of the trial, and managed the HeLP data collection with input from KW and other members of the study team. CA and SGD advised on the theoretical design of the intervention and the psychological and behavioural measures of change. SC wrote the statistical analysis plan, with analyses undertaken by AS and SC. MH supervised LP in the management of the physical activity data and its preparation for statistical analysis by AS and SC. JL, SC, and KW wrote the first draft of this paper and JL coordinated contributions from the other coauthors. All authors contributed to the writing and critical revision of the manuscript.

nd SC. MH supervised LP in the management of the physical activity data and its preparation for statistical analysis by AS and SC. JL, SC, and KW wrote the first draft of this paper and JL coordinated contributions from the other coauthors. All authors contributed to the writing and critical revision of the manuscript. Declaration of interests JL, SC, CG, SGD, MH, CA, RT, VP, RST, ER, LP, AS, and KW report grants from the Peninsula College of Medicine and Dentistry and non-financial methodological support during the transition from the exploratory trial to the definitive evaluation from the National Institute for Health Research (NIHR) Collaboration for Leadership in Applied Health, Research and Care (CLAHRC) for the South West Peninsula. SL reports grants from the CLAHRC for the South West Peninsula. SC, SL, CG, and SGD report grants from the NIHR. SGD reports personal fees from University College London and non-financial support from Knowledge Exchange conferences.

Introduction Children growing up in poverty are exposed to multiple psychological, physiological, and environmental risk factors that shape their development. Poverty increases exposure to poor sanitation and hygiene, acute and chronic infection, poor nutrition, food insecurity, abuse and neglect, and stress.1, 2 These conditions can have strong and enduring effects on child development across many domains.2 Globally, millions of children experience delays in physical health and cognitive development because of their exposures to poverty and related issues, such as nutrition, health care, education, and lack of stimulating environment.1 Programmes and policies reducing exposures to risk factors or enhancing protective factors can improve the trajectories of children's development. The approaches with the strongest evidence base so far include nutrition counselling, provision of fortified food (eg, lipid-based nutritional supplementation), micronutrient supplem-entation, and parenting support and education.3 Although these options have been tested, with many positive results, there is little agreement on the optimal design of programmes to maximise growth and improve cognitive or language development for children in low-income countries. It is also unknown to what extent combinations of these interventions might be additive or synergistic. Research in context Evidence before this study

Programmes and policies reducing exposures to risk factors or enhancing protective factors can improve the trajectories of children's development. The approaches with the strongest evidence base so far include nutrition counselling, provision of fortified food (eg, lipid-based nutritional supplementation), micronutrient supplem-entation, and parenting support and education.3 Although these options have been tested, with many positive results, there is little agreement on the optimal design of programmes to maximise growth and improve cognitive or language development for children in low-income countries. It is also unknown to what extent combinations of these interventions might be additive or synergistic. Research in context Evidence before this study Evidence linking health and nutrition with child development was scarce before the start of the WASH Benefits trial. Therefore, we did not do a systematic review before starting our trial; our evidence base at that time was the reviews by Walker et al, Grantham-McGregor et al, and Engle et al in The Lancet Series on Child Development in 2007 and 2011. Later on we identified more updated systematic reviews that examined associations between childhood health status and cognitive development in low-income and middle-income countries, the most recent of which was published in 2016 (Black et al). We updated the search in PubMed to July 15, 2017, using the search terms “water”, “sanitation”, and “child development”, for publications in English, and found an additional cohort study (Dearden et al, 2017) including several countries showing that children with access to improved water and toilet facilities in their first year of life had higher language scores (receptive vocabulary) at age 5 and 8 years. Overall, the methodological rigor of studies linking water, sanitation, and handwashing interventions and child development was poor, and only one randomised controlled trial had been done (Bowen et al, 2012). In this study, children younger than 2 years living in squatter settlements in urban Karachi, Pakistan, were randomly assigned to receive soap and intensive handwashing promotion for 9 months. Their scores on the Battelle Developmental Inventory II, 5 years after intervention at 5–7 years of age were 0·4 SDs higher than control children who received no intervention and no visits. We also reviewed evidence linking nutrition interventions and child development, and found that direct micronutrient supplementation to deficient populations led to improved child development (Walker et al, 2011). There was no evidence before this study about lipid-based nutrient supplements and child development, but while this study was ongoing, new evidence emerged that these supplements might be beneficial to child development in Bangladesh, Burkina Faso, and Ghana, but not in Malawi.

child development (Walker et al, 2011). There was no evidence before this study about lipid-based nutrient supplements and child development, but while this study was ongoing, new evidence emerged that these supplements might be beneficial to child development in Bangladesh, Burkina Faso, and Ghana, but not in Malawi. Added value of this study In this study, we assessed the single and combined effect of water, sanitation, handwashing, and nutrition interventions delivered with the support of frequent community worker visits on child development at age 1 and 2 years. We found some benefits in gross motor milestones when children were approximately 1 year old. We also found consistent developmental benefits in communication, gross motor, and personal social skills among children in all intervention groups when children were approximately 2 years old. Implications of all the available evidence Our findings suggest that interventions designed to improve water quality, sanitation, handwashing practices, or nutrition have cumulative beneficial effects on child development in addition to growth and reduced acute illness. Notably, we also recorded benefits of the single interventions in many cases, which would be a cheaper alternative to delivering combined interventions. Given that this trial was a test of efficacy, an important next step would be to consider how to bring these interventions to scale—both individually and combined—and to optimise effectiveness and cost-effectiveness.

interventions in many cases, which would be a cheaper alternative to delivering combined interventions. Given that this trial was a test of efficacy, an important next step would be to consider how to bring these interventions to scale—both individually and combined—and to optimise effectiveness and cost-effectiveness. Interventions targeting nutrition or development can be classified as development-specific interventions (ie, those addressing immediate determinants of nutrition and child development, such as inadequate nutrient intake and unsupportive caregiving practices), or development-sensitive interventions (ie, those addressing the underlying causes of undernutrition or poor development such as poverty, food insecurity, or scarcity of water and sanitation goods or services).4 Combined water, sanitation, and handwashing interventions fall into the broader category of nutrition-sensitive or development-sensitive interventions. The effects of these combined interventions on child development have not been extensively studied.

insecurity, or scarcity of water and sanitation goods or services).4 Combined water, sanitation, and handwashing interventions fall into the broader category of nutrition-sensitive or development-sensitive interventions. The effects of these combined interventions on child development have not been extensively studied. Exposure to poor water and sanitation causes diarrhoea in young children, and this is especially common in those living in low-income or middle-income countries.5 A recent meta-analysis examining effects of combined water, sanitation, and handwashing interventions on anthropometry showed small effects on improving linear growth in children younger than 5 years, and no effects on underweight or wasting.6 Findings of observational studies have shown associations between diarrhoea or other infectious diseases, and impaired cognitive outcomes.7, 8 One of the few randomised-controlled trials on this topic, done in Pakistan, found significant benefits of a handwashing promotion intervention on child development outcomes, but not on growth.9

vational studies have shown associations between diarrhoea or other infectious diseases, and impaired cognitive outcomes.7, 8 One of the few randomised-controlled trials on this topic, done in Pakistan, found significant benefits of a handwashing promotion intervention on child development outcomes, but not on growth.9 The WASH Benefits trial in Bangladesh was designed together with a WASH Benefits trial in Kenya to assess the independent and combined effects of water, sanitation, handwashing, and nutrition interventions on child growth, health, and development after 2 years of intervention.10 In Bangladesh, children receiving sanitation, handwashing, nutrition, and all combined interventions had less caregiver-reported diarrhoea, and children receiving interventions with nutritional components had modestly improved growth when compared with children from control households.11 In the companion trial in Kenya, small improvements in growth were seen in children receiving interventions with nutritional components, but none of the interventions reduced diarrhoea prevalence;12 some improvement in motor development was seen after 1 year, but there was no beneficial effect in measured child development indicators across all intervention groups after 2 years.10

ere seen in children receiving interventions with nutritional components, but none of the interventions reduced diarrhoea prevalence;12 some improvement in motor development was seen after 1 year, but there was no beneficial effect in measured child development indicators across all intervention groups after 2 years.10 The objective of this analysis was to assess whether: interventions improving water quality; sanitation; hand-washing with soap; water, sanitation, and handwashing in combination; nutrition; or water, sanitation, and hand-washing, and nutrition in combination would improve indicators of child development during the first 2 years of life; and whether the combination of water, sanitation, handwashing, and nutrition would improve child development measurements more than combined water, sanitation, and handwashing or nutrition alone.

d hand-washing, and nutrition in combination would improve indicators of child development during the first 2 years of life; and whether the combination of water, sanitation, handwashing, and nutrition would improve child development measurements more than combined water, sanitation, and handwashing or nutrition alone. Methods Study design Details on the study methods and rationale have been published previously.11, 13 The Bangladesh WASH Benefits study was a cluster-randomised controlled trial done in the rural villages of Gazipur, Kishoreganj, Mymensingh, and Tangail districts of central Bangladesh. We specifically chose areas with low ground water iron and arsenic (because these variables can affect the chlorine-based intervention), and where no major water, sanitation, or focused nutrition programmes were underway or planned by the government or large non-government organisations. We defined a cluster as consisting of eight pregnant women who lived close enough to each other so that a community promoter could readily walk to each compound. We maintained a 1 km buffer around each cluster to minimise the potential for spillover between clusters. The clusters were randomly assigned to seven study groups: drinking water treatment and safe storage; sanitation; handwashing; combined water, sanitation, and handwashing; nutrition; combined water, sanitation, handwashing, and nutrition; and a double-sized control group, which received no intervention or health promoter visits.

rs were randomly assigned to seven study groups: drinking water treatment and safe storage; sanitation; handwashing; combined water, sanitation, and handwashing; nutrition; combined water, sanitation, handwashing, and nutrition; and a double-sized control group, which received no intervention or health promoter visits. The study protocol was approved by human subjects committees at icddr,b (PR-11063), the University of California, Berkeley (UC Berkeley), CA, USA (2011-09-3652), and Stanford University, CA, USA (25863). Participants We enrolled pregnant women in their first or second trimester of pregnancy who intended to stay in their villages for 24 months post enrolment; their in utero children were considered to be the index children. After obtaining verbal and written informed consent, trained research assistants enrolled pregnant women in the study.

led pregnant women in their first or second trimester of pregnancy who intended to stay in their villages for 24 months post enrolment; their in utero children were considered to be the index children. After obtaining verbal and written informed consent, trained research assistants enrolled pregnant women in the study. Randomisation and masking Eight adjacent clusters formed a geographical block. An offsite investigator from UC Berkeley (BFA) used a random number generator to block randomise clusters into one of six intervention groups or into the double-sized control group, providing geographically pair-matched randomisation. We used a larger control group to improve the precision in comparing each of the six intervention groups against control. Participants and other community members were informed of their intervention group assignment after the baseline survey and randomisation. Interventions included distinct visible components and community health promoter visits, which meant that neither the study participants nor the data collectors could be masked to intervention assignment. Data collectors were different from the intervention teams.

assignment after the baseline survey and randomisation. Interventions included distinct visible components and community health promoter visits, which meant that neither the study participants nor the data collectors could be masked to intervention assignment. Data collectors were different from the intervention teams. Procedures Community health promoters who had completed 8 or more years of formal education, lived within walking distance of an intervention cluster, and who had social acceptance within the community were recruited to deliver the interventions. After qualifying for the job based on a written and oral exam, those who were accepted then attended a week long residential training session. The training sessions included basic training and component-specific training, as well as refresher training on a quarterly basis. Training programmes were interactive, included practice sessions and role play, and addressed technical components of the intervention, communication and negotiation skills. Community health promoters were instructed to make weekly visits for first 6 months, followed by fortnightly visits to intervention households, and were paid a monthly stipend (equivalent to US$20).

ice sessions and role play, and addressed technical components of the intervention, communication and negotiation skills. Community health promoters were instructed to make weekly visits for first 6 months, followed by fortnightly visits to intervention households, and were paid a monthly stipend (equivalent to US$20). The intervention was based on the Integrated Behavioural Model for Water Sanitation and Hygiene, addressing contextual, psychosocial, and technology factors at the societal and structural, community, interpersonal, individual, and habitual level.14 Key messages were delivered through Behaviour Change Communication materials—eg, visual aids including flip charts, posters, and reminder cue cards; interactive activities with songs and games; and the distribution of study group-specific hardware, products, or supplements. The water treatment group was provided with a 10 L vessel with a lid, tap, and regular supply of 33 mg sodium dichloroisocyanurate tablets (Medentech, Wexford, Ireland) per 10 L drinking water. We did spot checks for residual chlorine in the water treatment group from supplied water storage container by using a HACH colourimeter.

reatment group was provided with a 10 L vessel with a lid, tap, and regular supply of 33 mg sodium dichloroisocyanurate tablets (Medentech, Wexford, Ireland) per 10 L drinking water. We did spot checks for residual chlorine in the water treatment group from supplied water storage container by using a HACH colourimeter. The sanitation intervention was delivered at the compound level; compounds include rural households where patrilineally linked families live, which are usually arranged around a common courtyard. For all latrines in the compound that did not have a slab, a functional water seal or construction that prevented surface runoff of a faecal stream into the community were decommissioned and replaced. If the index household did not have their own latrine, the project built a latrine; the project also upgraded latrines for all the compound households that had an unhygienic latrine. The standard project intervention latrine was a double pit latrine with a water seal. The project also provided potties to children younger than 3 years and a sani-scoop (a spade-like hand tool) for removing faeces from the compound. The handwashing treatment group received two handwashing stations per index household, one with a 40 L water reservoir placed near the latrine and a 16 L reservoir for the kitchen. Each handwashing station included a basin to collect rinse water and a bottle with a regular supply of detergent sachets for making soapy water.

andwashing treatment group received two handwashing stations per index household, one with a 40 L water reservoir placed near the latrine and a 16 L reservoir for the kitchen. Each handwashing station included a basin to collect rinse water and a bottle with a regular supply of detergent sachets for making soapy water. The nutrition treatment group received a regular supply of lipid-based nutrient supplements (Nutriset, Malaunay, France) for children aged 6–24 months. Promoters instructed caregivers to feed a 10 g nutrient sachet (118 kcal, 9·6 g fat, 2·6 g protein, 12 vitamins, and ten minerals) to the index child twice daily; breastfeeding and complementary feeding intervention messages were adapted from the Alive and Thrive program in Bangladesh.15 Nutrition messages were focused on maternal extra food intake, dietary diversity, early initiation of colostrum within 30 min of the delivery, breastfeeding techniques (positioning and attachment), exclusive breastfeeding up to 6 months, and timely initiation of complementary feeding along with breastfeeding (6–24 months). Messages also included instructions about complementary food preparation along with serving portions and instructions about promoting children's self-feeding. By contrast, control children in the study were getting no such specific nutritional interventions or promoters visits. However, they might have received child feeding information (eg, about colostrum, breastfeeding, or complementary feeding) from government health workers during health visits or antenatal checkup as per national guidelines for infant and young child feeding in Bangladesh.

nutritional interventions or promoters visits. However, they might have received child feeding information (eg, about colostrum, breastfeeding, or complementary feeding) from government health workers during health visits or antenatal checkup as per national guidelines for infant and young child feeding in Bangladesh. After enrolment, baseline information was collected using standard questionnaires on the socioeconomic and demographic status of the participants, including parental education, maternal age, number of children younger than 18 years in household, total number of people in the compound, household assets and land ownership, homestead, water, sanitation, and hygiene conditions, behaviour of the household members, household food insecurity; and household structure (floor construction). At the baseline, year 1, and year 2 surveys, data were collected on intervention quality (chlorine spot check, compliance, child faeces disposal, presence of handwashing stations with soap and water, consumption rate of lipid-based nutrient supplements, and so on), as well as on maternal and child outcome measures, including health and hygiene information, morbidity, nutritional and developmental outcomes, and other biochemical measures.

, child faeces disposal, presence of handwashing stations with soap and water, consumption rate of lipid-based nutrient supplements, and so on), as well as on maternal and child outcome measures, including health and hygiene information, morbidity, nutritional and developmental outcomes, and other biochemical measures. Outcomes Here we report on the prespecified secondary child development outcomes of the trial.13 When children were aged about 1 year (range 7–17 months), we assessed gross motor milestones using the WHO module, which consists of direct assessments and parental reports of whether their child can perform certain actions (eg, standing with support).16 We also assessed language development (understanding and speaking words) using the MacArthur-Bates Communicative Development Inventories, which collects information about child language development via parental report; these well-established measures are validated for use in Bangladesh.17

(eg, standing with support).16 We also assessed language development (understanding and speaking words) using the MacArthur-Bates Communicative Development Inventories, which collects information about child language development via parental report; these well-established measures are validated for use in Bangladesh.17 When children were aged about 2 years (range 21–30 months), we assessed communication, gross motor, and personal social development using the Extended Ages and Stages Questionnaire (EASQ), which is mainly a parental report that we adapted for use in low-income and middle-income countries,18 using standard techniques.19 During adaptation, some direct tests (about 25% of total items) were added for behaviours that parents might fail to observe—eg, pointing at pictures in the picture book, naming body parts, kicking ball, offering toys to own mirror image, copying gestures (appendix). The domains of the tests were arranged in 2–3 month age bands with three responses: yes, sometimes, or not yet. To create the reference distributions for the communication, gross motor, and personal social subscales and the overall global scale of the EASQ, the summed age-specific raw scores from the control group were standardised with a mean of 0 and a SD of 1, yielding Z scores for each 2-month age band. Standardised Z scores for the rest of the sample were then created using the reference distribution for each age band.

scales and the overall global scale of the EASQ, the summed age-specific raw scores from the control group were standardised with a mean of 0 and a SD of 1, yielding Z scores for each 2-month age band. Standardised Z scores for the rest of the sample were then created using the reference distribution for each age band. At age 2 years, we used two executive function tests—the A-not-B task and the Tower test—to directly assess children's impulse control, ability to initiate action, ability to sustain attention, and their persistence. The A-not-B task focused on working memory of children and was recently adapted and standardised in Bangladesh for young children.20 In this test a treat is hidden in front of a child in one of two shallow wells on a wooden board and covered with the two opaque cups; after distraction for 5 s, the child is asked to lift the right cup and get the treat. Sum of correct attempts out of ten trials gives the total score. The Tower test, where children took turns to make towers out of eight blocks with a tester (university graduate), was adapted after extensive piloting. In pilot tests on 77 children younger than 2 years, the Tower test showed good test–retest reliability (correlation coefficient r=0·73; p=0·02) and moderate concurrent validity, when measured by assessing correlation with the Family Care Indicator at the same time (r=0·21, p=0·05).

was adapted after extensive piloting. In pilot tests on 77 children younger than 2 years, the Tower test showed good test–retest reliability (correlation coefficient r=0·73; p=0·02) and moderate concurrent validity, when measured by assessing correlation with the Family Care Indicator at the same time (r=0·21, p=0·05). At ages 1 and 2 years, we collected information about behaviours related to responsive parenting (eg, activities and outings for children, toys and books available in the home), using items adapted from the Home Measurement for Observation for the Environment (HOME)21 and from the UNICEF Multi-Indicator Cluster Surveys.22 Maternal depressive symptoms were assessed at both timepoints using the Centers for Epidemiological Studies-Depression Scale (CESD), a brief, widely used measure of 20 statements that assess the likelihood of depressive symptomology; this tool has been previously culturally adapted and used in other studies set in Bangladesh.23 All tests of child development and maternal wellbeing were piloted on 77 non-study children aged 18–24 months and their mothers, and showed face validity, meaningful correlation with developmental measures, association with sociodemographic variables, and test–retest reliability at 7-day intervals (correlation coefficient r for all tests >0·70).

ent and maternal wellbeing were piloted on 77 non-study children aged 18–24 months and their mothers, and showed face validity, meaningful correlation with developmental measures, association with sociodemographic variables, and test–retest reliability at 7-day intervals (correlation coefficient r for all tests >0·70). Eight testers (university graduates) received 5–10 days of extensive hands-on training, including theoretical and practical sessions. When agreement achieved at least 90% between testers and trainers, the interviewers were considered ready for the main study. For 5–10% of total tests we looked for ongoing reliability across the time period by comparing the testers with a trained supervisor (gold standard) who had a psychology background. We arranged refresher trainings every 6 months or when correlation coefficients (r) for any tester with a supervisor fell below 0·85.

udy. For 5–10% of total tests we looked for ongoing reliability across the time period by comparing the testers with a trained supervisor (gold standard) who had a psychology background. We arranged refresher trainings every 6 months or when correlation coefficients (r) for any tester with a supervisor fell below 0·85. Statistical analysis The trial was powered to detect a difference of 0·15 in the primary outcome length-for-age Z score (LAZ) in comparisons of intervention groups against control, accounting for repeated measures within clusters.13 This design allocated one third of clusters to intervention (for any single group comparison against control), assumed seven children per cluster, 80% power with a two-sided alpha of 0·05, and intra-cluster correlation of 0·05, and had a minimum detectable effect size for the EASQ Z scores of 0·16 using a standard equation for cluster randomised trials.24 Under the same assumptions, the minimum detectable effect for comparison of combined versus single intervention groups for the EASQ Z scores was 0·18.

and intra-cluster correlation of 0·05, and had a minimum detectable effect size for the EASQ Z scores of 0·16 using a standard equation for cluster randomised trials.24 Under the same assumptions, the minimum detectable effect for comparison of combined versus single intervention groups for the EASQ Z scores was 0·18. All analyses were intention to treat. Because randomisation was geographically pair-matched in blocks of eight clusters, we estimated unadjusted mean differences using generalised linear models that considered pair matching and block-level clustering. The geographical pair-matching ensured that measurement timing was balanced across groups. The pair-matched analyses removed any potential confounding from seasonal changes in baseline risk. For each comparison, we estimated p values using a paired t-test and cluster-level means. Because the children were of wide age ranges and the tests are age dependent, we used age-specific z scores of tests, within 2 month age bands, and controlled for age in the analysis.

unding from seasonal changes in baseline risk. For each comparison, we estimated p values using a paired t-test and cluster-level means. Because the children were of wide age ranges and the tests are age dependent, we used age-specific z scores of tests, within 2 month age bands, and controlled for age in the analysis. We summarised the distribution of continuous outcomes using Gaussian kernel density smoothers using default bandwidth and kernel selection in R. We also estimated adjusted mean differences by adjusting for baseline covariates that could be considered potential confounders: child sex, maternal age, maternal height (in cm), parents' education in year of schooling, number of children younger than 18 years in household, total number of people in the compound, food insecurity of household (measured using Household Hunger Scale), housing materials (construction materials and utilities), household assets, distance to water source, and month of measurement. For adjusted analyses, we only included the covariates that were associated (p<0·2) with the outcomes using a likelihood ratio test. We conducted pre-specified subgroup analyses to examine effect modification by child sex, maternal parity, age and education, household hunger score and socioeconomic status.

measurement. For adjusted analyses, we only included the covariates that were associated (p<0·2) with the outcomes using a likelihood ratio test. We conducted pre-specified subgroup analyses to examine effect modification by child sex, maternal parity, age and education, household hunger score and socioeconomic status. To compare attainment rates for each of the WHO motor milestones, we estimated hazard ratios from current status data using a semiparametric generalised additive model with complementary log-log link and baseline hazard fit with a monotonic cubic spline. All analyses were done with R (version 3.2.4) and STATA (version 13.0). The trial is registered with ClinicalTrials.gov, NCT01590095. Role of the funding source The funder reviewed and approved the experimental design, but was not involved in data collection, data analysis, data interpretation, or writing of the report. The corresponding author had full access to all of the data in the study and had final responsibility for the decision to submit for publication.

unding source The funder reviewed and approved the experimental design, but was not involved in data collection, data analysis, data interpretation, or writing of the report. The corresponding author had full access to all of the data in the study and had final responsibility for the decision to submit for publication. Results We identified 13 279 pregnant women in their first or second trimester. About half of these women were excluded to create 1 km buffer zones between intervention clusters. Between May 31, 2012, and July 7, 2013, we randomly allocated 720 clusters and enrolled 5551 women to one of the six intervention groups (90 clusters each) or to the control group (180 clusters; figure 1). Index children of 928 (17%) enrolled women were lost to follow-up in year 1 and an additional 201 (3%) women in year 2 for various reasons—eg, stillbirth (n=363 [6%]), death before the final assessment (n=220 [4%]), out migration (n=98 [2%]), absent on repeated follow-up (n=397 [7%]), or refusal (n=52 [<1%]). Losses to follow-up were balanced across groups (appendix). 4757 children were assessed at 1 year and 4403 at 2 years (figure 1). Treatment groups were well balanced on all baseline demographic characteristics, including maternal and paternal education, household composition, and household wealth; groups were also balanced with regards to drinking water source, sanitation practices and access to sanitation-related supplies, and handwashing supplies and practices (table 1). There was high adherence of all groups to the assigned interventions, with uptake of more than 80% in the single intervention groups and similar uptake in combined intervention groups (appendix).Figure 1 Trial profile

practices and access to sanitation-related supplies, and handwashing supplies and practices (table 1). There was high adherence of all groups to the assigned interventions, with uptake of more than 80% in the single intervention groups and similar uptake in combined intervention groups (appendix).Figure 1 Trial profile Numbers are children except where specified. Attrition was only at the child level; no cluster dropped out. WASH=water, sanitation, and handwashing. EASQ=Extended Ages and Stages Questionnaire. Table 1 Baseline characteristics

practices and access to sanitation-related supplies, and handwashing supplies and practices (table 1). There was high adherence of all groups to the assigned interventions, with uptake of more than 80% in the single intervention groups and similar uptake in combined intervention groups (appendix).Figure 1 Trial profile Numbers are children except where specified. Attrition was only at the child level; no cluster dropped out. WASH=water, sanitation, and handwashing. EASQ=Extended Ages and Stages Questionnaire. Table 1 Baseline characteristics Control group (N=1382) Water group (N=698) Sanitation group (N=696) Handwashing group (N=688) Combined water, sanitation, and handwashing group (N=702) Nutrition (N=699) Combined water, sanitation, handwashing, and nutrition group (N=686) Maternal Age (years) 23·6 (5·0) 23·7 (5·2) 23·7 (5·2) 23·8 (5·5) 24·3 (5·5) 23·7 (5·1) 23·8 (5·5) Years of education 5·9 (3·4) 5·8 (3·4) 5·8 (3·5) 5·8 (3·3) 5·9 (3·3) 5·8 (3·5) 5·6 (3·5) Paternal Years of education 4·9 (4·0) 4·9 (4·1) 5·0 (4·2) 4·6 (4·1) 5·0 (4·2) 4·8 (4·0) 4·7 (3·9) Works in agriculture 414 (30%) 224 (32%) 204 (29%) 249 (36%) 216 (31%) 232 (33%) 207 (30%) Household Number of people 4·7 (2·3) 4·6 (2·2) 4·7 (2·1) 4·7 (2·2) 4·7 (2·1) 4·7 (2·2) 4·7 (2·1) Has electricity 784 (57%) 422 (60%) 408 (59%) 405 (59%) 426 (61%) 409 (59%) 412 (60%) Has a cement floor 145 (10%) 82 (12%) 85 (12%) 55 (8%) 77 (11%) 67 (10%) 72 (10%) Acres of agricultural land owned 0·15 (0·21) 0·14 (0·20) 0·14 (0·22) 0·14 (0·20) 0·15 (0·23) 0·16 (0·27) 0·14 (0·38) Drinking water Tubewell primary water source 1038 (75%) 500 (72%) 519 (75%) 482 (70%) 546 (78%) 519 (74%) 504 (73%) Stored water observed at home 666 (48%) 353 (51%) 341 (49%) 347 (50%) 304 (43%) 301 (43%) 331 (48%) Sanitation Daily defecation in the open Adult men 97 (7%) 39 (6%) 52 (8%) 64 (9%) 54 (8%) 59 (9%) 50 (7%) Adult women 62 (4%) 18 (3%) 33 (5%) 31 (5%) 29 (4%) 39 (6%) 24 (4%) Children aged 8 to <15 years 53 (10%) 25 (9%) 28 (9%) 43 (15%) 30 (10%) 23 (8%) 28 (10%) Children aged 3 to <8 years 267 (38%) 141 (37%) 137 (38%) 137 (39%) 137 (38%) 129 (39%) 134 (37%) Children aged 0 to <3 years 245 (82%) 112 (85%) 117 (84%) 120 (85%) 123 (79%) 128 (85%) 123 (88%) Latrine Owned 750 (54%) 363 (52%) 374 (54%) 372 (54%) 373 (53%) 377 (54%) 367 (53%) Concrete slab 1251 (95%) 644 (95%) 610 (92%) 613 (93%) 620 (93%) 620 (94%) 621 (94%) Functional water seal 358 (31%) 183 (31%) 177 (30%) 162 (28%) 152 (26%) 183 (31%) 155 (27%) Visible stool on slab or floor 625 (48%) 350 (53%) 332 (52%) 335 (52%) 289 (44%) 331 (51%) 298 (46%) Owned a potty 61 (4%) 27 (4%) 28 (4%) 35 (5%) 27 (4%) 36 (5%) 30 (4%) Human faeces observed In the house 114 (8%) 65 (9%) 56 (8%) 70 (10%) 48 (7%) 58 (8%) 49 (7%) In

%) 177 (30%) 162 (28%) 152 (26%) 183 (31%) 155 (27%) Visible stool on slab or floor 625 (48%) 350 (53%) 332 (52%) 335 (52%) 289 (44%) 331 (51%) 298 (46%) Owned a potty 61 (4%) 27 (4%) 28 (4%) 35 (5%) 27 (4%) 36 (5%) 30 (4%) Human faeces observed In the house 114 (8%) 65 (9%) 56 (8%) 70 (10%) 48 (7%) 58 (8%) 49 (7%) In child's play area 21 (2%) 6 (1%) 6 (1%) 8 (1%) 7 (1%) 8 (1%) 7 (1%) Handwashing Within six steps of latrine Has water 178 (14%) 83 (13%) 81 (13%) 63 (10%) 67 (10%) 62 (10%) 72 (11%) Has soap 88 (7%) 50 (8%) 48 (8%) 34 (5%) 42 (7%) 32 (5%) 36 (6%) Within six steps of kitchen Has water 118 (9%) 51 (8%) 51 (8%) 45 (7%) 61 (9%) 61 (9%) 60 (9%) Has soap 33 (3%) 18 (3%) 14 (2%) 13 (2%) 15 (2%) 23 (3%) 18 (3%) Nutrition Household is food secure* 932 (67%) 495 (71%) 475 (68%) 475 (69%) 482 (69%) 479 (69%) 485 (71%) Data are n (%) or mean (SD). Percentages were estimated from slightly smaller denominators than those shown at the top of the table for the following variables due to missing values: father works in agriculture, open defecation, latrine has a concrete slab, latrine has a functional water seal, visible stool on latrine slab or floor, ownership of child potty, observed faeces in the house or child's play area, handwashing variables. * Assessed by the Household Food Insecurity Access Scale.25

Percentages were estimated from slightly smaller denominators than those shown at the top of the table for the following variables due to missing values: father works in agriculture, open defecation, latrine has a concrete slab, latrine has a functional water seal, visible stool on latrine slab or floor, ownership of child potty, observed faeces in the house or child's play area, handwashing variables. * Assessed by the Household Food Insecurity Access Scale.25 At 1 year follow-up, the median age of the children was 11 months (IQR 10–13). The age of attainment of each of the motor milestones in the study group was slightly delayed compared with the WHO reference population (table 2). There were some improvements in motor milestone attainment for children in the nutrition group or the combined water, sanitation, handwashing, and nutrition group (table 3). Specifically, compared with the control group, the combined water, sanitation, handwashing, and nutrition group had a greater rate of attaining the standing alone milestone (conditional on not standing upright on both for 10 s; hazard ratio [HR], 95% CI 1·01–1·40) and the nutrition group had a greater rate of attaining the walking alone milestone (conditional on not walking independently for 5 steps; 1·32, 1·07–1·62). The rate of attaining the walking alone milestone was higher in the combined water, sanitation, handwashing, and nutrition group than in the combined group without nutrition (HR 1·29, 95% CI 1 ·01–1·65). These improvements were robust to the inclusion of covariates in adjusted analyses (appendix).Table 2 Estimated age of attainment for each of the motor milestones among children in the study population compared with the WHO reference population

than in the combined group without nutrition (HR 1·29, 95% CI 1 ·01–1·65). These improvements were robust to the inclusion of covariates in adjusted analyses (appendix).Table 2 Estimated age of attainment for each of the motor milestones among children in the study population compared with the WHO reference population WHO growth standards reference*, median (IQR) age of attainment, months WASH Benefits study children†, median (IQR) age of attainment, months Age 7–9 months (N=282) Age 9–11 months (N=1603) Age 11–13 months (N=1989) Age 13–15 months (N=843) Sitting without support 5·9 (5·8–6·0) ·· 280 (99·6%) 1588 (99·3%) 1979 (99·5%) 835 (99·3%) Standing with assistance 7·4 (6·6–8·4) 8·7 (8·0–9·3) 141 (50·5%) 1320 (83·1%) 1860 (94·5%) 814 (97·4%) Hands-and-knees crawling 8·3 (8·2–8·4) ·· 162 (57·7%) 1111 (69·5%) 1504 (76·0%) 672 (80·0%) Walking with assistance 9·0 (8·2–10·0) 10·5 (9·1–11·3) 29 (10·4%) 739 (46·4%) 1579 (80·3%) 760 (91·1%) Standing alone 10·8 (9·7–12·0) 11·9 (10·9–13·7) 6 (2·1%) 237 (14·9%) 965 (48·9%) 631 (75·8%) Walking alone 12·0 (11·0–13·0) 13·0 (11·9–14·1) 1 (0·4%) 47 (2·9%) 527 (26·7%) 529 (63·3%) * Published data from the WHO Multicentre Growth Reference Study.14 † Estimated using non-parametric maximum likelihood estimator of cumulative probability of attainment. Median age could not be estimated for the sitting without support and hands-and-knees crawling milestones because >80% of children had already achieved this milestone before the assessment. Table 3 Relative rate of motor milestone attainment after 1 year of intervention

† Estimated using non-parametric maximum likelihood estimator of cumulative probability of attainment. Median age could not be estimated for the sitting without support and hands-and-knees crawling milestones because >80% of children had already achieved this milestone before the assessment. Table 3 Relative rate of motor milestone attainment after 1 year of intervention N Hazard ratio vs control group (95% CI) Hazard ratio vs combined water, sanitation, and handwashing group group (95% CI) Hazard ratio vs nutrition group (95% CI) Standing with assistance Control 1161 1 (ref) ·· ·· Water 598 0·99 (0·86–1·14) ·· ·· Sanitation 593 1·03 (0·89–1·19) ·· ·· Handwashing 584 0·90 (0·78–1·03) ·· ·· Combined water, sanitation, handwashing 594 0·95 (0·83–1·09) 1 (ref) ·· Nutrition 580 1·13 (0·98–1·31) ·· 1 (ref) Combined water, sanitation, handwashing, and nutrition 603 1·05 (0·91–1·20) 1·11 (0·95–1·30) 0·94 (0·80–1·10) Hands-and-knees crawling Control 1171 1 (ref) ·· ·· Water 601 0·94 (0·83–1·06) ·· ·· Sanitation 593 0·98 (0·86–1·11) ·· ·· Handwashing 589 0·85 (0·75–0·96) ·· ·· Combined water, sanitation, handwashing 595 0·91 (0·80–1·03) 1 (ref) ·· Nutrition 584 1·02 (0·90–1·15) ·· 1 (ref) Combined water, sanitation, handwashing, and nutrition 606 0·98 (0·87–1·11) 1·08 (0·94–1·25) 0·96 (0·84–1·11) Walking with assistance Control 1160 1 (ref) ·· ·· Water 598 1·06 (0·93–1·21) ·· ·· Sanitation 591 1·09 (0·95–1·25) ·· ·· Handwashing 583 0·87 (0·76–1·01) ·· ·· Combined water, sanitation, handwashing 594 0·98 (0·85–1·13) 1 (ref) ·· Nutrition 580 1·12 (0·98–1·29) ·· 1 (ref) Combined water, sanitation, handwashing and nutrition 604 0·99 (0·87–1·14) 1·02 (0·87–1·19) 0·88 (0·75–1·03) Standing alone Control 1165 1 (ref) ·· ·· Water 600 1·14 (0·97–1·34) ·· ·· Sanitation 591 1·10 (0·93–1·29) ·· ·· Handwashing 585 0·96 (0·81–1·14) ·· ·· Combined water, sanitation, handwashing 592 1·02 (0·86–1·21) 1 (ref) ·· Nutrition 582 1·14 (0·96–1·34) ·· 1 (ref) Combined water, sanitation, handwashing, and nutrition 607 1·19 (1·01–1·40) 1·17 (0·97–1·41) 1·03 (0·86–1·25) Walking alone Control 1165 1 (ref) ·· ·· Water 600 1·26 (1·03–1·54) ·· ·· Sanitation 592 1·15 (0·93–1·41) ·· ·· Handwashing 585 1·09 (0·87–1·35) ·· ·· Combined water, sanitation, handwashing 593 0·94 (0·75–1·17) 1 (ref) ·· Nutrition 581 1·32 (1·07–1·62) ·· 1 (ref) Combined water, sanitation, handwashing, and nutrition 607 1·21 (0·98–1·48) 1·29 (1·01–1·65) 0·90 (0·72–1·14)

6 (1·03–1·54) ·· ·· Sanitation 592 1·15 (0·93–1·41) ·· ·· Handwashing 585 1·09 (0·87–1·35) ·· ·· Combined water, sanitation, handwashing 593 0·94 (0·75–1·17) 1 (ref) ·· Nutrition 581 1·32 (1·07–1·62) ·· 1 (ref) Combined water, sanitation, handwashing, and nutrition 607 1·21 (0·98–1·48) 1·29 (1·01–1·65) 0·90 (0·72–1·14) At year 2 follow-up, the median age of children was 26 months (IQR 24–27). There were consistent beneficial effects of the intervention on all sub-scales of the EASQ child development measure (communication, gross motor, personal social), with largest effects in the combined water, sanitation, handwashing, and nutrition group (table 4, figure 2, appendix). Compared with the control group, all intervention groups except the water treatment group had higher scores on the communication and gross motor subscales, and all intervention groups had higher scores on the personal social subscale (table 4). In the combined EASQ measure including all subscales, effect sizes were smallest in the water treatment group (mean difference 0·15, 95% CI 0·04–0·26 compared with control) and largest in the combined water, sanitation, handwashing, and nutrition group (0·37, 0·27–0·46; figure 2). The improvements were robust to the inclusion of covariates in adjusted analyses (appendix).Figure 2 Kernal density plots of combined Z scores of the Extended Ages and Stages Questionnaire at year 2 follow-up

l) and largest in the combined water, sanitation, handwashing, and nutrition group (0·37, 0·27–0·46; figure 2). The improvements were robust to the inclusion of covariates in adjusted analyses (appendix).Figure 2 Kernal density plots of combined Z scores of the Extended Ages and Stages Questionnaire at year 2 follow-up Kernel density plots summarise the distribution of combined Z scores of index children who were born into the study and were between 21–30 months (median 26 months, IQR 24–27) at the time of measurement. Data are mean difference (95% CI). WASH=water, sanitation, and handwashing. N=nutrition. Table 4 Standardised differences in scores on the communication, gross motor, personal social, and combined scales of the Extended Ages and Stages Questionnaire after 2 years of intervention

Kernel density plots summarise the distribution of combined Z scores of index children who were born into the study and were between 21–30 months (median 26 months, IQR 24–27) at the time of measurement. Data are mean difference (95% CI). WASH=water, sanitation, and handwashing. N=nutrition. Table 4 Standardised differences in scores on the communication, gross motor, personal social, and combined scales of the Extended Ages and Stages Questionnaire after 2 years of intervention N, mean (SD) Mean difference vs control group (95% CI) Mean difference vs combined water, sanitation, and handwashing group (95% CI) Mean difference vs nutrition group (95% CI) Communication Z score Control 1099, 0·00 (1·00) 0 (ref) ·· ·· Water 565, 0·09 (0·96) 0·10 (−0·02 to 0·21) ·· ·· Sanitation 555, 0·20 (0·95) 0·21 (0·09 to 0·33) ·· ·· Handwashing 543, 0·20 (0·94) 0·21 (0·10 to 0·31) ·· ·· Combined water, sanitation, and handwashing 547, 0·15 (0·93) 0·14 (0·03 to 0·26) 0 (ref) ·· Nutrition 541, 0·19 (0·86) 0·19 (0·10 to 0·28) ·· 0 (ref) Combined water, sanitation, handwashing, and nutrition 553, 0·25 (0·90) 0·26 (0·16 to 0·36) 0·11 (−0·02 to 0·25) 0·07 (−0·05 to 0·19) Gross motor Z score Control 1099, 0·00 (1·00) 0 (ref) ·· ·· Water 557, 0·01 (0·93) 0·01 (−0·11 to 0·13) ·· ·· Sanitation 549, 0·12 (0·93) 0·12 (0·00 to 0·25) ·· ·· Handwashing 535, 0·10 (0·93) 0·12 (0·00 to 0·23) ·· ·· Combined water, sanitation, and handwashing 539, 0·16 (0·87) 0·16 (0·04 to 0·27) 0 (ref) ·· Nutrition 528, 0·18 (0·95) 0·19 (0·08 to 0·30) ·· 0 (ref) Combined water, sanitation, handwashing, and nutrition 546, 0·14 (0·94) 0·14 (0·03 to 0·25) −0·01 (−0·12 to 0·11) −0·05 (−0·19 to 0·10) Personal social Z score Control 1099, 0·00 (1·00) 0 (ref) ·· ·· Water 557, 0·11 (0·91) 0·13 (0·01 to 0·25) ·· ·· Sanitation 544, 0·28 (0·97) 0·29 (0·18 to 0·40) ·· ·· Handwashing 528, 0·26 (0·94) 0·28 (0·17 to 0·40) ·· ·· Combined water, sanitation, and handwashing 538, 0·28 (1·01) 0·27 (0·16 to 0·38) 0 (ref) ·· Nutrition 528, 0·22 (0·97) 0·22 (0·11 to 0·33) ·· 0 (ref) Combined water, sanitation, handwashing, and nutrition 538, 0·34 (0·98) 0·35 (0·24 to 0·46) 0·07 (−0·06 to 0·20) 0·13 (−0·01 to 0·28) Combined Z score Control 1099, 0·00 (1·00) 0 (ref) ·· ·· Water 539, 0·14 (0·86) 0·15 (0·04 to 0·26) ·· ·· Sanitation 531, 0·31 (0·86) 0·31 (0·19 to 0·43) ·· ·· Handwashing 502, 0·27 (0·87) 0·29 (0·19 to 0·39) ·· ·· Combined water, sanitation, and handwashing 519, 0·25 (0·90) 0·24 (0·14 to 0·35) 0 (ref) ·· Nutrition 506, 0·27 (0·83) 0·28 (0·18 to 0·37) ·· 0 (ref) Combined water, sanitation, handwashing, and nutrition 522, 0·36 (0·81) 0·37 (0·27 to 0·46) 0·12 (−0·01 to 0·24) 0·10 (−0·02 to 0·21)

·27 (0·87) 0·29 (0·19 to 0·39) ·· ·· Combined water, sanitation, and handwashing 519, 0·25 (0·90) 0·24 (0·14 to 0·35) 0 (ref) ·· Nutrition 506, 0·27 (0·83) 0·28 (0·18 to 0·37) ·· 0 (ref) Combined water, sanitation, handwashing, and nutrition 522, 0·36 (0·81) 0·37 (0·27 to 0·46) 0·12 (−0·01 to 0·24) 0·10 (−0·02 to 0·21) We noted benefits in the comprehension subscale of the MacArthur Communicative Development Inventories for the water treatment group, handwashing group, and in both combined intervention groups in year 1 (appendix), as well as in both the comprehension and the expressive language subscales for all intervention groups in year 2 (table 5). However, the effect sizes were not consistently significantly different across the intervention groups. There were no effects of any intervention on executive function measures (table 5).Table 5 Effect of the interventions on communication and executive functions after 2 years

l intervention groups in year 2 (table 5). However, the effect sizes were not consistently significantly different across the intervention groups. There were no effects of any intervention on executive function measures (table 5).Table 5 Effect of the interventions on communication and executive functions after 2 years N, mean (SD) Mean difference vs control group (95% CI) Mean difference vs combined water, sanitation, and handwashing group (95% CI) Mean difference vs nutrition group (95% CI) MacArthur-Bates Communicative Development Inventories Comprehension score Control 1106, 0·00 (1·00) 0 (ref) ·· ·· Water 535, 0·18 (0·88) 0·20 (0·08 to 0·32) ·· ·· Sanitation 524, 0·16 (0·87) 0·18 (0·05 to 0·31) ·· ·· Handwashing 519, 0·21 (0·80) 0·22 (0·11 to 0·33) ·· ·· Water, sanitation, and handwashing 513, 0·18 (0·85) 0·18 (0·07 to 0·29) 0 (ref) ·· Nutrition 501, 0·19 (0·80) 0·19 (0·08 to 0·29) ·· 0 (ref) Water, sanitation, handwashing, and nutrition 528, 0·25 (0·86) 0·26 (0·14 to 0·38) 0·07 (−0·05 to 0·19) 0·07 (−0·05 to 0·20) Expressive language score Control 1106, 0·00 (1·00) 0 (ref) ·· ·· Water 497, 0·18 (0·88) 0·18 (0·07 to 0·30) ·· ·· Sanitation 473, 0·17 (0·90) 0·17 (0·06 to 0·29) ·· ·· Handwashing 472, 0·19 (0·86) 0·19 (0·09 to 0·30) ·· ·· Water, sanitation, and handwashing 470, 0·13 (0·97) 0·11 (0·00 to 0·23) 0 (ref) ·· Nutrition 461, 0·19 (0·86) 0·18 (0·07 to 0·29) ·· 0 (ref) Water, sanitation, handwashing, and nutrition 484, 0·19 (0·92) 0·20 (0·08 to 0·32) 0·07 (−0·07 to 0·22) 0·01 (−0·14 to 0·16) Executive function Tower test Z score Control 1106, 0·00 (1·00) 0 (ref) ·· ·· Water 582, 0·09 (0·99) 0·09 (−0·01 to 0·20) ·· ·· Sanitation 564, 0·01 (1·00) 0·02 (−0·11 to 0·15) ·· ·· Handwashing 559, 0·12 (0·90) 0·13 (0·02 to 0·24) ·· ·· Water, sanitation, and handwashing 566, 0·06 (1·01) 0·06 (−0·05 to 0·18) 0 (ref) ·· Nutrition 550, 0·07 (0·94) 0·07 (−0·04 to 0·18) ·· 0 (ref) Water, sanitation, handwashing, and nutrition 571, 0·11 (0·95) 0·10 (−0·01 to 0·22) 0·03 (−0·11 to 0·16) 0·02 (−0·12 to 0·15) A-not-B test Z score Control 1106, 0·00 (1·00) 0 (ref) ·· ·· Water 579, 0·17 (0·88) 0·17 (0·06 to 0·28) ·· ·· Sanitation 560, 0·10 (0·92) 0·10 (−0·01 to 0·20) ·· ·· Handwashing 555, 0·08 (0·94) 0·08 (−0·02 to 0·18) ·· ·· Water, sanitation, and handwashing 565, 0·05 (0·98) 0·05 (−0·05 to 0·16) 0 (ref) ·· Nutrition 545, 0·04 (0·95) 0·04 (−0·06 to 0·14) ·· 0 (ref) Water, sanitation, handwashing, and nutrition 568, 0·08 (0·93) 0·07 (−0·01 to 0·16) 0·01 (−0·10 to 0·13) 0·03 (−0·07 to 0·14

dwashing 555, 0·08 (0·94) 0·08 (−0·02 to 0·18) ·· ·· Water, sanitation, and handwashing 565, 0·05 (0·98) 0·05 (−0·05 to 0·16) 0 (ref) ·· Nutrition 545, 0·04 (0·95) 0·04 (−0·06 to 0·14) ·· 0 (ref) Water, sanitation, handwashing, and nutrition 568, 0·08 (0·93) 0·07 (−0·01 to 0·16) 0·01 (−0·10 to 0·13) 0·03 (−0·07 to 0·14 ) Measures of maternal depressive symptoms were lower in all intervention groups than the control group at both year 1 and year 2 (appendix). Compared with the control group, responsive parenting assessed by the HOME scales was improved at year 1 in the sanitation group, handwashing group, and combined water, sanitation, handwashing, and nutrition group), and at year 2 in the water treatment group, nutrition group, and combined water, sanitation, handwashing, and nutrition group (appendix). We found no consistent evidence for effect modification by child sex; maternal parity, age, education, household hunger score, and socioeconomic status in additional pre-specified subgroup analyses (appendix).

Measures of maternal depressive symptoms were lower in all intervention groups than the control group at both year 1 and year 2 (appendix). Compared with the control group, responsive parenting assessed by the HOME scales was improved at year 1 in the sanitation group, handwashing group, and combined water, sanitation, handwashing, and nutrition group), and at year 2 in the water treatment group, nutrition group, and combined water, sanitation, handwashing, and nutrition group (appendix). We found no consistent evidence for effect modification by child sex; maternal parity, age, education, household hunger score, and socioeconomic status in additional pre-specified subgroup analyses (appendix). Discussion In this trial of independent and combined water, sanitation, handwashing, and nutritional interventions provided to households in rural Bangladesh, we found some benefits to children after 1 year of exposure when they were about 12 months old. We also found consistent developmental benefits across a range of outcomes among children in all intervention groups after 2 years of exposure when children were approximately 24 months old; there were no significant effects on measures of executive function. Rates of motor milestone attainment at age 1 year were faster in children in the combined water, sanitation, handwashing, and nutrition group and in the nutrition alone group. Benefits in domain-specific developmental outcomes (communication, gross motor, and personal social subscales of the EASQ; MacArthur-Bates comprehension and expressive language) measured at age 2 years were more consistent, and were apparent in almost all intervention groups; the only domain in which there was no consistent effect was executive function at this age.

es (communication, gross motor, and personal social subscales of the EASQ; MacArthur-Bates comprehension and expressive language) measured at age 2 years were more consistent, and were apparent in almost all intervention groups; the only domain in which there was no consistent effect was executive function at this age. The distribution plots suggest some potential ceiling effects in the MacArthur-Bates communication (both comprehensive and expressive language) scores in year 1 but not in year 2. This finding could be due to reporting bias of mothers about their children's early abilities, which reduced with progression of age in year 2 (appendix). The two executive tests we used were possibly slightly difficult for these 2-year-old Bangladeshi children. Although executive function skills begin to develop shortly after birth, different tests to measure these abilities suggest that they are best assessed between 3 to 5 years of age, when substantial brain growth in areas responsible for these skills occurs.26 For language, gross motor, personal social, and all communication domains, effect sizes ranged from 0·13 to 0·35 SDs.

tly after birth, different tests to measure these abilities suggest that they are best assessed between 3 to 5 years of age, when substantial brain growth in areas responsible for these skills occurs.26 For language, gross motor, personal social, and all communication domains, effect sizes ranged from 0·13 to 0·35 SDs. Several development-specific and development-sensitive mechanisms might have played a role in affecting children's development in our study.27 Possible mechanisms linking improved water, sanitation, and handwashing practices with better developmental outcomes include reduced infection, reduced inflammation, and increased social interaction, all of which could improve synaptic connections as well as myelination in the CNS, thus benefitting developmental outcomes.28 A main link connecting water, sanitation, and handwashing interventions with child development is care practices of parents and other adult caregivers, relating to feeding, nutrition, and health, which can have direct effects on child outcomes.29 However, the improved developmental outcomes could also have resulted from an indirect pathway by improving maternal wellbeing and reducing maternal stress or depression. If the interventions themselves change a mother's ability to provide support and nurturing care for her children, maternal depression may connect water, sanitation, and handwashing interventions and improved cognitive outcomes.30 We found clear effects of all intervention groups on reducing maternal depressive symptoms, and some evidence for an improved home environment, suggesting that this pathway might be operating here. One of our key hypotheses for why the interventions were effective is that families received frequent visits and support from community health workers in all intervention groups.

ucing maternal depressive symptoms, and some evidence for an improved home environment, suggesting that this pathway might be operating here. One of our key hypotheses for why the interventions were effective is that families received frequent visits and support from community health workers in all intervention groups. Another clear pathway connecting water, sanitation, and handwashing interventions with developmental outcomes is via nutrition, particularly through linear growth faltering (low height-for-age or stunting);31 however, this pathway is unlikely to have affected our results, given that the water, sanitation, and handwashing interventions—alone or in combination—did not affect linear growth.11 In the nutrition group, which included lipid-based nutrient supplements and nutritional messages, there were improvements on stunting and wasting in index children,11 suggesting that some effects on development could occur via growth, or directly from the lipid-based nutrient supplement formulation. In a recent study evaluating prenatal and postnatal lipid-based nutrient supplements and micronutrient powders,20 lipid-based nutrient supplements had positive effects on motor and language development, but no effects on personal social behaviour or executive function. Similarly, a study in Burkina Faso32 also reported significant benefits of lipid-based nutrient supplements on children's motor, language, and personal social development, and a trial in Ghana33 found significant beneficial effects of lipid-based nutrient supplements on children's development. By contrast, there was no benefit of lipid-based nutrient supplements on child development measures in Malawian children given the supplement at 6–18 months of age.34

d personal social development, and a trial in Ghana33 found significant beneficial effects of lipid-based nutrient supplements on children's development. By contrast, there was no benefit of lipid-based nutrient supplements on child development measures in Malawian children given the supplement at 6–18 months of age.34 Another pathway that potentially links water, sanitation, and handwashing interventions with developmental outcomes is via enteric infections, including intestinal worms caused by poor sanitation, which can then result in iron deficiency and have negative consequences for child development.35 Experiencing repeated and prolonged episodes of diarrhoea could have direct effects on cognitive outcomes.36 In our trial, children receiving sanitation, handwashing, nutrition, and all combined interventions had roughly 40% lower diarrhoea prevalence compared with control.11 The intervention groups might also have reduced systemic inflammation (eg, CRP, sCD14, IL-1β and IL6), which has been associated with improved neurodevelopmental outcomes,37 and this pathway may also be operating here. Assessment is currently underway to investigate this potential pathway.13

ce compared with control.11 The intervention groups might also have reduced systemic inflammation (eg, CRP, sCD14, IL-1β and IL6), which has been associated with improved neurodevelopmental outcomes,37 and this pathway may also be operating here. Assessment is currently underway to investigate this potential pathway.13 In the parallel WASH Benefits study in rural Kenya, there were few developmental benefits despite similar types of interventions and similar growth benefits in the two nutrition groups.10, 12 This discrepancy could be due to less intense contact between health promoters with respondents (one contact per month in Kenya vs up to six contacts per month in Bangladesh), leading to less uptake of the targeted behaviours. An important area for future research is to establish whether similar effects, as found in Bangladesh, can be achieved with a frequency of promoter contacts that could be supported in large-scale interventions.

h in Kenya vs up to six contacts per month in Bangladesh), leading to less uptake of the targeted behaviours. An important area for future research is to establish whether similar effects, as found in Bangladesh, can be achieved with a frequency of promoter contacts that could be supported in large-scale interventions. The main strengths of this study include the pair-matched cluster-randomised design, large sample size, high intervention adherence (as determined by objective indicators11), intensive interventions (often weekly or more frequently), and the use of multiple direct and indirect developmental indicators that were locally adapted to Bangladeshi children. Data collectors were rigorously trained for at least 1 week and showed high inter-rater agreement and test–retest reliabilities before the start of data collection. Ongoing quality assurance was also monitored. The consistent pattern of benefits in most of the domains of development across intervention groups further strengthens the internal consistency of the results.

t 1 week and showed high inter-rater agreement and test–retest reliabilities before the start of data collection. Ongoing quality assurance was also monitored. The consistent pattern of benefits in most of the domains of development across intervention groups further strengthens the internal consistency of the results. Despite these strengths, the inferences that we can draw from this study are limited by some features of the design. One limitation is that we cannot know how much benefit resulted from social interaction and how much from reduced disease, reduced inflammation, or improved nutrition. Interventions delivered at the household level were intense (delivered weekly for the first 6 months and then fortnightly for next 18 months) and started from the second trimester of pregnancy. Thus, future research of this size and magnitude should explore direct and indirect effects of the interventions on outcomes, accounting for mediation by maternal depression, indicators of resources at home such as toys and books, and parental behaviours, along with active and passive control groups to account for contact with community health workers. Some of the questions answered by parental report (eg, MacArthur-Bates Communicative Development Inventories or some of the EASQ) might have been affected by courtesy bias, and in our study the parents and data collectors were not masked. To minimise the possibility of courtesy bias, we used a non-traditional approach to develop promoter skills that focused on collaborative problem-solving, rather than forceful advice-giving. Also, the assessment team was different from promoter team; the assessment team was unaware of the intervention details, joined the study midway, and only visited the families once a year to collect developmental data. Another limitation of our study is that effect estimates for single versus combined interventions fell slightly below the minimum detectable effect given the design, and thus were not statistically significant. Finally, it is unclear what predictive validity these early tests of child development might have for longer-term improvements in the trajectory of children's intellectual development. Thus, an important next step would be to follow up the children for longer to assess their development over time, especially into the pre-school and schooling years using direct child assessment measures.

child development might have for longer-term improvements in the trajectory of children's intellectual development. Thus, an important next step would be to follow up the children for longer to assess their development over time, especially into the pre-school and schooling years using direct child assessment measures. In conclusion, our findings suggest that interventions designed to improve water quality, sanitation, handwashing practices, or nutrition have cumulative beneficial effects on children's developmental outcomes and go beyond the traditionally measured outcomes of growth and acute illness. Although some of our findings suggest that a combined intervention might have advantages over the individual interventions, we also found benefits of the single interventions in many cases, which would be a cheaper alternative to delivering combined interventions. Finally, given that this trial was a test of efficacy, an important next step would be to consider how to bring these interventions to scale—both individually and combined—with a focus on optimising intervention effectiveness and cost-effectiveness. Key learnings from the Scaling Up Nutrition movement, for example, have been that larger scale-up efforts require a clear vision for change, an enabling contextual environment (eg, household, community, political), actors, stakeholders and champions at multiple levels, government ownership, multiple incentives, adequate financial resources, and frameworks for monitoring, evaluation, learning, and accountability.38 Supplementary Material Supplementary appendix

In conclusion, our findings suggest that interventions designed to improve water quality, sanitation, handwashing practices, or nutrition have cumulative beneficial effects on children's developmental outcomes and go beyond the traditionally measured outcomes of growth and acute illness. Although some of our findings suggest that a combined intervention might have advantages over the individual interventions, we also found benefits of the single interventions in many cases, which would be a cheaper alternative to delivering combined interventions. Finally, given that this trial was a test of efficacy, an important next step would be to consider how to bring these interventions to scale—both individually and combined—with a focus on optimising intervention effectiveness and cost-effectiveness. Key learnings from the Scaling Up Nutrition movement, for example, have been that larger scale-up efforts require a clear vision for change, an enabling contextual environment (eg, household, community, political), actors, stakeholders and champions at multiple levels, government ownership, multiple incentives, adequate financial resources, and frameworks for monitoring, evaluation, learning, and accountability.38 Supplementary Material Supplementary appendix Acknowledgments The research protocol was funded by Global Development grant OPPGD759 from the Bill & Melinda Gates Foundation to the University of California, Berkeley (UC Berkeley) to International Centre for Diarrhoeal Disease Research, Bangladesh (icddr,b). icddr,b acknowledges with gratitude the commitment of Bill & Melinda Gates Foundation and the UC Berkeley, to its research efforts. icddr,b is also grateful to the Governments of Bangladesh, Canada, Sweden, and the UK for providing core and unrestricted support. We also offer our sincere gratitude to the study participants who participated in the trial and community health promoters, field workers, and supervisors who delivered the interventions.

. icddr,b is also grateful to the Governments of Bangladesh, Canada, Sweden, and the UK for providing core and unrestricted support. We also offer our sincere gratitude to the study participants who participated in the trial and community health promoters, field workers, and supervisors who delivered the interventions. Contributors FT and LCHF drafted the manuscript with input from all co-authors. FT, LCHF, PK, and KKJ developed the child development assessment protocols, and piloted and refined the instruments. SPL, PJW, MR, and LU developed the water, sanitation, and handwashing intervention, and TA and CPS developed the nutrition interventions. MR, LU, and SA oversaw piloting and subsequent study implementation, contributed to refinements in interventions and measurements, and ensured validity of measurements. KKD, BFA, and JMC developed the analytical approach, conducted the statistical analysis, and constructed the tables and figures. All authors have read, contributed to, and approved the final version of the manuscript. Declaration of interests We declare no competing interests.

Introduction Mortality from diarrhoea has decreased since the 1990s, but it still causes up to 0·6 million deaths annually in children younger than 5 years, with most deaths occurring in Africa and southeast Asia.1, 2 A range of interventions might have contributed to this reduction, such as safe water and hygiene, handwashing, exclusive breastfeeding, measles vaccine, improvement in use of oral rehydration salts,3 and zinc supplementation.4 Routine rotavirus vaccination is expected to further reduce diarrhoeal severity, morbidity, and mortality, although recent studies have shown that bacterial pathogens, such as non-typhoidal Salmonella, Cryptosporidium, Shigella, and Escherichia coli, are important causes of diarrhoeal deaths within hospitals.5 As a result, there is a global call to prioritise examination of risk factors for continued diarrhoeal mortality and investigate delivery of proven interventions.6

l pathogens, such as non-typhoidal Salmonella, Cryptosporidium, Shigella, and Escherichia coli, are important causes of diarrhoeal deaths within hospitals.5 As a result, there is a global call to prioritise examination of risk factors for continued diarrhoeal mortality and investigate delivery of proven interventions.6 Understanding the presenting clinical features that identify children at risk of death is important for front-line clinicians who often solely rely on clinical signs to prioritise immediate care. Standardised WHO guidance7, 8 on management of diarrhoea with dehydration, which recommends fluid treatment, is available and its usage can modify the association between certain clinical risk factors and mortality. Therefore, examination of risk factors for mortality should also investigate any moderating effect of treatment. We investigate clinical risk factors for in-hospital death and risk modification associated with intended use of WHO fluid treatment guidance in children with diarrhoea and dehydration admitted to 13 hospitals in Kenya.9 We use a large routine dataset, based on documentation of clinical practice outside specific research settings, with variation likely in compliance with clinical guidance and contextual factors (captured at the hospital level). Research in context Evidence before this study

Understanding the presenting clinical features that identify children at risk of death is important for front-line clinicians who often solely rely on clinical signs to prioritise immediate care. Standardised WHO guidance7, 8 on management of diarrhoea with dehydration, which recommends fluid treatment, is available and its usage can modify the association between certain clinical risk factors and mortality. Therefore, examination of risk factors for mortality should also investigate any moderating effect of treatment. We investigate clinical risk factors for in-hospital death and risk modification associated with intended use of WHO fluid treatment guidance in children with diarrhoea and dehydration admitted to 13 hospitals in Kenya.9 We use a large routine dataset, based on documentation of clinical practice outside specific research settings, with variation likely in compliance with clinical guidance and contextual factors (captured at the hospital level). Research in context Evidence before this study Diarrhoea still causes up to 0·6 million deaths annually in children younger than 5 years. Research into the optimisation of delivery and scaling up of existing interventions are thought to be the most urgent priorities to further reduce mortality. However, identification of risk factors for diarrhoeal deaths is also among the top ten research priorities for the reduction of diarrhoeal mortality. To identify studies of risk factors for mortality from diarrhoea in children (younger than 18 years), we searched PubMed for studies published in English between database inception and June 31, 2017. We used various combinations of the following search terms: “diarrhoea”, “diarrhea”, “risk”, “odds”, “mortality”, and “death”. We also perused the bibliographies of retrieved articles. We found case-control studies that reported on risk factors of diarrhoeal deaths in children in resource-poor settings but found none that used routine clinical data. No studies have investigated the effect of use of recommended rehydration guidance on mortality in routine clinical practice. Most available studies have been done in the context of investigating microbial aetiology of diarrhoea rather than clinical characteristics, which are what front-line clinicians use for making treatment decisions in resource-poor settings.

e of recommended rehydration guidance on mortality in routine clinical practice. Most available studies have been done in the context of investigating microbial aetiology of diarrhoea rather than clinical characteristics, which are what front-line clinicians use for making treatment decisions in resource-poor settings. Added value of this study This study shows that that children with other signs of severe illness (suggestive of comorbidity), and not signs of dehydration alone, are most at risk of in-hospital death from diarrhoea and dehydration. Our findings also show that comorbidities are common and that correct fluid prescription is associated with reduced risk of death in these patients. Implications of all the available evidence These findings highlight the need for further studies on how to manage diarrhoea and dehydration complicated with comorbidities. This research is especially important given that different fluid management approaches could be recommended for children with various comorbidities. We also show the benefit of adherence to recommended practice at the hospital level. This study also raises questions regarding the need to investigate appropriate strategies to optimise delivery of recommended guidance in routine settings in hospitals within resource-poor settings, which often have inadequately trained clinical personnel.

fit of adherence to recommended practice at the hospital level. This study also raises questions regarding the need to investigate appropriate strategies to optimise delivery of recommended guidance in routine settings in hospitals within resource-poor settings, which often have inadequately trained clinical personnel. Methods Study design and participants For this observational, association study, we analysed prospective routine clinical information on admission, immediate treatment, and discharge, which was collected from 13 first referral-level Kenyan hospitals that constitute the Clinical Information Network (CIN). We excluded one CIN facility from this analysis because it is contextually different from the other CIN facilities; it is staffed by non-physician clinicians (ie, non-degree trained clinicians) and is a health centre with inpatient beds rather than being a first-referral hospital. A detailed description of CIN facilities has been previously published.

this analysis because it is contextually different from the other CIN facilities; it is staffed by non-physician clinicians (ie, non-degree trained clinicians) and is a health centre with inpatient beds rather than being a first-referral hospital. A detailed description of CIN facilities has been previously published. We screened the database for children meeting the following criteria: age 1–59 months (as guidelines only apply to this group), diarrhoea as a presenting symptom, and a full dataset available. We excluded children with only a minimal dataset or severe acute malnutrition because children with severe acute malnutrition have different fluid treatment guidelines.8 Then we identified children with diarrhoea as a presenting symptom or diagnosis or with dehydration as a diagnosis from the full dataset. Within this group, we included in our analyses only children with both diarrhoea as a presenting complaint or diagnosis and also a primary or secondary diagnosis of dehydration (some, severe, shock, or unclassified). These criteria defined a population eligible for treatment according to WHO and Kenyan guidelines for diarrhoea and dehydration. Diarrhoea was considered only a presenting symptom rather than a diagnosis in children classified as having no dehydration on the standard admission form. Data for HIV status were not recorded comprehensively in this population, but the analysis included children with a diagnosis of HIV. Maternal HIV prevalence in Kenya is 6% and the cumulative 5 year mother-to-child transmission rate is 15%; as such, we expect only a small proportion of children in the dataset to have undiagnosed HIV infection.10

t recorded comprehensively in this population, but the analysis included children with a diagnosis of HIV. Maternal HIV prevalence in Kenya is 6% and the cumulative 5 year mother-to-child transmission rate is 15%; as such, we expect only a small proportion of children in the dataset to have undiagnosed HIV infection.10 The Kenya Medical Research Institute (KEMRI) Scientific and Ethical Review Committee approved the CIN study enabling use of de-identified data without individual patient consent.

t recorded comprehensively in this population, but the analysis included children with a diagnosis of HIV. Maternal HIV prevalence in Kenya is 6% and the cumulative 5 year mother-to-child transmission rate is 15%; as such, we expect only a small proportion of children in the dataset to have undiagnosed HIV infection.10 The Kenya Medical Research Institute (KEMRI) Scientific and Ethical Review Committee approved the CIN study enabling use of de-identified data without individual patient consent. Procedures The hospitals use WHO and locally adapted guidance for management of common conditions.8, 9, 10, 11 In brief, these hospitals have implemented two clinical data collection tools (standard paediatric admission records and discharge forms) and have a dedicated data clerk who enters information about admission, treatment, and discharge, once the patient is discharged, into a non-proprietary electronic tool.12, 13 The clerks are trained and regularly updated on how to abstract data from medical notes and treatment sheets, including fluid prescription sheets, and on how to interpret them. Error checks are done before the data are uploaded and synchronised into a central server, in which further quality checks are done. Any discrepancies noted at this stage are raised with respective clerks who reconcile them. Periodic visits are made to the participating hospitals by the data management team who re-enter a number of randomly selected files to ascertain the accuracy of data entered by the clerks. A minimal dataset, which consists of data required for the routine health information system from all admissions (patient age, sex, diagnoses, and outcome), is collected for a random sample of otherwise eligible admissions in two high-volume hospitals (to reduce the data entry workload) and in all 13 hospitals for surgical or burns cases, admissions younger than 1 month, and admissions during periods when the single clerk is on leave. The randomisation sequence is system generated automatically and not within the clerks' control. A full dataset on clinical presentation, diagnoses, treatments, and outcomes is collected on all other cases and at all other times. We aimed to investigate clinical signs associated with mortality and whether prescription of recommended fluid guidance is associated with a reduced risk of mortality.

he clerks' control. A full dataset on clinical presentation, diagnoses, treatments, and outcomes is collected on all other cases and at all other times. We aimed to investigate clinical signs associated with mortality and whether prescription of recommended fluid guidance is associated with a reduced risk of mortality. Outcomes We aimed to examine clinical risk factors for in-hospital death and risk modification associated with intended use of WHO fluid treatment guidance in children admitted with diarrhoea and dehydration across the hospitals. Data analysis Patient characteristics examined include sex, age (≤12 months or >12 months), duration of diarrhoea (≤14 days or >14 days), length of illness (≤2 days or >2 days), history of bloody diarrhoea, malaria status, and abnormal signs obtained on examination of various systems organised as airway, breathing (respiratory system), circulation, hydration status, and disability (neurological system). A child was deemed to have an abnormal system if any sign within the specific system was abnormal (see panel 1 for definitions of abnormal signs). In the case of dehydration, we also created a variable to represent children with clinical signs indicative of severe dehydration (defined in Kenyan guidelines as the presence of both sunken eyes and delayed skin pinch). Correct fluid prescription was defined based on WHO and Kenyan guidance (panel 2).Panel 1 Definition of terms Airway signs Abnormal airway signs* (only stridor analysed this study) . Circulatory signs

Data analysis Patient characteristics examined include sex, age (≤12 months or >12 months), duration of diarrhoea (≤14 days or >14 days), length of illness (≤2 days or >2 days), history of bloody diarrhoea, malaria status, and abnormal signs obtained on examination of various systems organised as airway, breathing (respiratory system), circulation, hydration status, and disability (neurological system). A child was deemed to have an abnormal system if any sign within the specific system was abnormal (see panel 1 for definitions of abnormal signs). In the case of dehydration, we also created a variable to represent children with clinical signs indicative of severe dehydration (defined in Kenyan guidelines as the presence of both sunken eyes and delayed skin pinch). Correct fluid prescription was defined based on WHO and Kenyan guidance (panel 2).Panel 1 Definition of terms Airway signs Abnormal airway signs* (only stridor analysed this study) . Circulatory signs Presence of any one or more of the following: capillary refill time greater than 2 s (delayed capillary refill time), temperature gradient (cold hands and feet), weak pulse volume, or pallor. Comorbidity Dehydration plus any of the following: malaria, pneumonia, HIV, tuberculosis, anaemia, meningitis, rickets, or asthma. Dehydration signs Presence of either delayed skin pinch (greater than 1 s) or sunken eyes, or both. Impaired circulation Presence of any one or more of the following: weak pulse volume, temperature gradient (skin temperature up to shoulder or elbow), or capillary refill time longer than 2 s. Impaired consciousness

Dehydration plus any of the following: malaria, pneumonia, HIV, tuberculosis, anaemia, meningitis, rickets, or asthma. Dehydration signs Presence of either delayed skin pinch (greater than 1 s) or sunken eyes, or both. Impaired circulation Presence of any one or more of the following: weak pulse volume, temperature gradient (skin temperature up to shoulder or elbow), or capillary refill time longer than 2 s. Impaired consciousness AVPU (Alert, Voice, Pain, Unresponsive) score less than A. Malaria endemic zone A hospital was regarded as located in a high malaria endemic zone if malaria diagnoses comprised more than 50% of admission diagnoses. Neurological or disability signs Presence of any one or more of the following: convulsions, neck stiffness, bulging anterior fontanelle, inability to drink or breastfeed, or impaired consciousness. Pneumonia History of cough or difficulty breathing, age older than 60 days, lower chest wall indrawing or tachypnoea-non-severe pneumonia, and any danger sign (oxygen saturation <90%, cyanosis, inability to drink or breastfeed, impaired consciousness, or grunting). Respiratory illness signs Presence of any one or more of grunting, tachypnoea, chest indrawing, acidotic breathing, crackles, or crepitations. Severe acute malnutrition Defined as clinical diagnosis of severe acute malnutrition, mid-upper arm circumference less than 11·5 cm, or weight for height Z score less than −3 standard deviations. Clinical shock Clinical diagnosis of shock made by clinician or fluid bolus given. Tachypnoea

Presence of any one or more of grunting, tachypnoea, chest indrawing, acidotic breathing, crackles, or crepitations. Severe acute malnutrition Defined as clinical diagnosis of severe acute malnutrition, mid-upper arm circumference less than 11·5 cm, or weight for height Z score less than −3 standard deviations. Clinical shock Clinical diagnosis of shock made by clinician or fluid bolus given. Tachypnoea Respiratory rate more than 50 breaths per min if aged 12 months or younger, or more than 40 breaths per min if older than 12 months. WHO shock Presence of all of an AVPU score less than A, weak pulse, and capillary refill time longer than 3 s in the presence of diagnosis of dehydration. Panel 2 Definitions of correct fluid prescription based on WHO and Kenyan guidance In this study, either WHO plan B, WHO plan C, or shock management7, 8 were correctly prescribed. Plan B was correct when given to children classified as having some dehydration and who had not been prescribed bolus fluid and received oral fluid (prescribed for a duration of 4 h or prescribed to be given at regular intervals) or prescribed intravenous fluid (not plan C or fluid bolus and volume <200 mL/kg for a duration of 24 h) plus oral fluid. Plan C was correct when given to children with a diagnosis of severe dehydration, who had not been prescribed bolus, and in whom oral fluid was used. Correct volume of plan C was 30 mL/kg for step 1 and 100 mL/kg for step 2. The correct duration of plan C was 6 h or less.

Plan B was correct when given to children classified as having some dehydration and who had not been prescribed bolus fluid and received oral fluid (prescribed for a duration of 4 h or prescribed to be given at regular intervals) or prescribed intravenous fluid (not plan C or fluid bolus and volume <200 mL/kg for a duration of 24 h) plus oral fluid. Plan C was correct when given to children with a diagnosis of severe dehydration, who had not been prescribed bolus, and in whom oral fluid was used. Correct volume of plan C was 30 mL/kg for step 1 and 100 mL/kg for step 2. The correct duration of plan C was 6 h or less. Shock management was correct in children indicated as having shock and prescribed fluid bolus plus correct plan C. Intravenous fluid or bolus was correct if normal (0·9%) saline or Ringer's lactate was used.

Plan C was correct when given to children with a diagnosis of severe dehydration, who had not been prescribed bolus, and in whom oral fluid was used. Correct volume of plan C was 30 mL/kg for step 1 and 100 mL/kg for step 2. The correct duration of plan C was 6 h or less. Shock management was correct in children indicated as having shock and prescribed fluid bolus plus correct plan C. Intravenous fluid or bolus was correct if normal (0·9%) saline or Ringer's lactate was used. We did multiple imputation using chained equations to deal with missing data. Using Stata version 15.1, we did imputation with 100 iterations to produce ten imputed datasets on the assumption that co-variable data were missing at random.14, 15 The imputation model included all variables to be considered as risk factors, auxiliary variables (fever, history of vomiting, and cough), and outcomes, but we excluded any variable with greater than 30% missingness.16 We studied risk factors for overall in-hospital mortality and early (within 2 days) in-hospital mortality using mixed-effects logistic regression models, with patient-level data (level I) nested within hospitals (level II) and hospital location in malaria zone as a level II fixed effect. Univariable models (unadjusted) were fitted on the imputed datasets for each patient characteristic and malaria zone location as fixed effects, and hospital intercept as a random effect. The multivariable model (model II; adjusted) was constructed using all patient characteristics and a backward variable selection procedure with a p value for exclusion of 0·05, while maintaining the same multilevel structure as the univariable model. A priori, we decided to include age, sex, and malaria diagnosis in the adjusted model. Final model estimates were derived using Rubin rules.14 Final risk factors are variables independently associated with outcome in the multivariable model (Wald test p value <0·05). Because our analysis focuses on a population in whom we have excluded no dehydration, we also investigated the association between signs identified as risk factors in model II and mortality in a broader population of children admitted with diarrhoea using the same approach as done for those with diarrhoea and dehydration to investigate the generalisability of findings.

m we have excluded no dehydration, we also investigated the association between signs identified as risk factors in model II and mortality in a broader population of children admitted with diarrhoea using the same approach as done for those with diarrhoea and dehydration to investigate the generalisability of findings. We investigated effect modification of admission fluid prescription on risk of death in children with signs of dehydration, abnormal respiratory signs, impaired circulation, anaemia, and abnormal neurological signs, by including a binary term for correct initial fluid prescription in the model for early death (within 2 days from admission). We calculated the relative excess odds due to interaction, the attributable proportion, and the multiplicative interaction odds ratio (OR).17 Interactions were analysed in the final model (model II) one at a time. Analysis for effect modification was restricted to early deaths because we hypothesised that this is the group whose outcome might be affected by fluid management at admission. Our database collected only fluid prescribed at admission. Patients with no information on fluid management were excluded from the analysis for effect modification. Data sharing statement Data for this report are under the primary jurisdiction of the Ministry of Health in Kenya. Enquiries about using the data can be made to the KEMRI-Wellcome Trust Research Programme Data Governance Committee.

We investigated effect modification of admission fluid prescription on risk of death in children with signs of dehydration, abnormal respiratory signs, impaired circulation, anaemia, and abnormal neurological signs, by including a binary term for correct initial fluid prescription in the model for early death (within 2 days from admission). We calculated the relative excess odds due to interaction, the attributable proportion, and the multiplicative interaction odds ratio (OR).17 Interactions were analysed in the final model (model II) one at a time. Analysis for effect modification was restricted to early deaths because we hypothesised that this is the group whose outcome might be affected by fluid management at admission. Our database collected only fluid prescribed at admission. Patients with no information on fluid management were excluded from the analysis for effect modification. Data sharing statement Data for this report are under the primary jurisdiction of the Ministry of Health in Kenya. Enquiries about using the data can be made to the KEMRI-Wellcome Trust Research Programme Data Governance Committee. Role of the funding source The funders of the study had no role in study design, data collection, data analysis, data interpretation, or writing of this manuscript. SA, PA, DG, AA, GI, KS, and ME had access to the raw data. The corresponding author had full access to all the data in the study and had final responsibility for the decision to submit for publication.

study had no role in study design, data collection, data analysis, data interpretation, or writing of this manuscript. SA, PA, DG, AA, GI, KS, and ME had access to the raw data. The corresponding author had full access to all the data in the study and had final responsibility for the decision to submit for publication. Results Between Oct 1, 2013, and Dec 1, 2016, 19 839 (36%) of 55 048 eligible paediatric patients with a full dataset admitted to the 13 CIN hospitals had diarrhoea or dehydration. 8562 patients had both diarrhoea and dehydration and 9618 had diarrhoea as a symptom only (without dehydration; figure 1). We did the analysis to determine risk factors and effect modification with fluid therapy in children who had both diarrhoea and dehydration (n=8562).Figure 1 Study profile CIN=Clinical Information Network.

Results Between Oct 1, 2013, and Dec 1, 2016, 19 839 (36%) of 55 048 eligible paediatric patients with a full dataset admitted to the 13 CIN hospitals had diarrhoea or dehydration. 8562 patients had both diarrhoea and dehydration and 9618 had diarrhoea as a symptom only (without dehydration; figure 1). We did the analysis to determine risk factors and effect modification with fluid therapy in children who had both diarrhoea and dehydration (n=8562).Figure 1 Study profile CIN=Clinical Information Network. Among those with diarrhoea and dehydration (n=8562), 5160 (60%) had some dehydration, 2434 (28%) had severe dehydration, 431 (5%) had shock, and 537 (6%) had no classification for the degree of dehydration (table 1). The overall mortality was 9% (759 of 8562) and diarrhoea and dehydration together were associated with 28% (759 of 2711) of all deaths in eligible children in the full dataset. Case fatality in children with some dehydration was 5% (240 of 5160), severe dehydration was 13% (310 of 2434), shock was 42% (179 of 431), and in unclassified cases was 6% (30 of 537). 205 (27%) of the 759 deaths occurred within the first day and a total of 486 (64%) had died within 2 days; median duration to death was 1 day (IQR 0–2). At baseline, fever (6153 [76%] of 8067) and vomiting (6780 [83%] of 8169) were common. Although only 37 (<1%) of 8562 patients fulfilled the WHO criteria for shock (panel 1), a clinical diagnosis of shock was present in 537 (6%) of 8562 patients. Most participants (7184 [84%] of 8562) had a second diagnosis (comorbidity), and these included, among others, slide-positive malaria (2766 [32%]), pneumonia (3065 [36%]), anaemia (428 [5%]), and possible meningitis (642 [8%]). Median ages for admitted patients was 12·0 months (IQR 8·0–18·0).Table 1 Clinical characteristics at admission of patients with diarrhoea and dehydration

(comorbidity), and these included, among others, slide-positive malaria (2766 [32%]), pneumonia (3065 [36%]), anaemia (428 [5%]), and possible meningitis (642 [8%]). Median ages for admitted patients was 12·0 months (IQR 8·0–18·0).Table 1 Clinical characteristics at admission of patients with diarrhoea and dehydration Complete dataset Missing data for variable (n=8562) Mean proportion with characteristic present in imputed data (n= 85 620) Total Characteristic present Demographics Girls 8485 3724 (47%) 77 (1%) 37 574 (44%) Age ≤12 months 8562 4941 (58%) 0 49 410 (58%) History Length of illness >2 days 8104 5353 (66%) 458 (5%) 56 292 (66%) Diarrhoea >14 days 7465 252 (3%) 1097 (13%) 2843 (3%) Diarrhoea bloody 7585 277 (4%) 977 (11%) 3174 (4%) Airway Stridor 7453 97 (1%) 1109 (13%) 1147 (1%) Airway signs 7453 97 (1%) 1109 (13%) 1147 (1%) Breathing or respiratory Tachypnoea 6342 1972 (31%) 2220 (26%) 25 570 (30%) Grunting 7687 637 (8%) 875 (10%) 7038 (8%) Indrawing 7780 1621 (21%) 782 (9%) 17 185 (20%) Crackles or crepitations 7863 1193 (15%) 699 (8%) 12 860 (15%) Respiratory signs 8161 3141 (39%) 401 (5%) 35 973 (42%) Circulation Capillary refill >2 s 6404 705 (11%) 2158 (25%) 9481 (11%) Temperature gradient 6270 655 (10%) 2292 (27%) 8921 (11%) Weak pulse volume 7381 946 (13%) 1181 (14%) 11 050 (13%) Circulatory signs 7697 1641 (21%) 865 (10%) 20 079 (23%) Pallor 7954 1240 (16%) 608 (7%) 13 392 (16%) Dehydration Delayed skin pinch 7562 3569 (47%) 1000 (12%) 40 113 (47%) Sunken eyes 7511 3875 (52%) 1051 (12%) 42 843 (50%) Dehydrations signs 7889 4998 (63%) 673 (8%) 54 025 (63%) Dehydrations signs (severe) 7889 2446 (31%) 673 (8%) 28 931 (34%) Disability or neurological Convulsions 7836 856 (11%) 726 (9%) 9332 (11%) Inability to drink or breastfeed 7533 1864 (25%) 1029 (12%) 20 906 (25%) Impaired consciousness (AVPU<A) 7974 878 (11%) 588 (7%) 9453 (11%) Neurological signs 8118 2263 (28%) 444 (5%) 24 773 (29%) Others Malaria 8562 2771 (32%) 0 27 710 (32%) Death 8562 759 (9%) 0 7590 (9%) Data are n (%) unless stated otherwise. AVPU=Alert, Voice, Pain, Unresponsive.

7533 1864 (25%) 1029 (12%) 20 906 (25%) Impaired consciousness (AVPU<A) 7974 878 (11%) 588 (7%) 9453 (11%) Neurological signs 8118 2263 (28%) 444 (5%) 24 773 (29%) Others Malaria 8562 2771 (32%) 0 27 710 (32%) Death 8562 759 (9%) 0 7590 (9%) Data are n (%) unless stated otherwise. AVPU=Alert, Voice, Pain, Unresponsive. In 537 (6%) of 8562 patients, information on fluid prescribed was missing. Correct fluid (fluid type, volume, use of rate on infusion recommended for age of child) was prescribed in 3760 (44%) of 8025 participants. Among children with information on fluid prescribed, WHO plan B was correctly prescribed to 3398 (66%) of 5160 participants, WHO plan C to 331 (14%) of 2434 participants, and shock fluid prescriptions to 31 (7%) of 431 participants. 3569 (45%) of 8025 participants received at least some intravenous fluid whereas the rest of the participants received oral fluids only. Because intravenous fluids were more likely to be given to patients with more severe forms of dehydration, we included use of intravenous fluids in our analysis for risk factors as a proxy for disease severity. Antibacterials were prescribed in 5250 (61%) of 8562 participants, antimalarials in 2621 (32%) of 8252 participants, and any antimicrobial (antibacterial or antimalarial) in 6321 (74%) of 8562 participants. Electrolyte testing was generally unavailable across the hospitals.9

isk factors as a proxy for disease severity. Antibacterials were prescribed in 5250 (61%) of 8562 participants, antimalarials in 2621 (32%) of 8252 participants, and any antimicrobial (antibacterial or antimalarial) in 6321 (74%) of 8562 participants. Electrolyte testing was generally unavailable across the hospitals.9 All variables included in the analysis had less than 27% of missing values (table 1), but a multilevel model using complete cases would have incorporated only 24% of cases (at least one missing variable). The population characteristics in the imputed data were similar to the characteristics before imputation (table 1). Analysis of the association of each covariable with mortality in models using imputed data suggested that female sex, age of 12 months or younger, length of illness of more than 2 days, diarrhoea duration of more than 14 days, abnormal airway signs, abnormal respiratory signs, abnormal circulatory signs, pallor, signs of dehydration, use of intravenous fluids (proxy for severity), and abnormal neurological signs were all significantly associated with in-hospital death across hospitals, whereas bloody diarrhoea and malaria parasitaemia were not (figure 2). Backward selection excluded bloody diarrhoea; as such, it was not included in the final multivariable model.Figure 2 Risk factors for in-hospital mortality in children with diarrhoea and dehydration

ssociated with in-hospital death across hospitals, whereas bloody diarrhoea and malaria parasitaemia were not (figure 2). Backward selection excluded bloody diarrhoea; as such, it was not included in the final multivariable model.Figure 2 Risk factors for in-hospital mortality in children with diarrhoea and dehydration (A) Association of each covariable with in-hospital mortality in models using imputed data. (B) Association with in-hospital mortality after adjustment for all patient-level covariables. *Odds ratios not calculated. †Proxy measure for illness severity.

ssociated with in-hospital death across hospitals, whereas bloody diarrhoea and malaria parasitaemia were not (figure 2). Backward selection excluded bloody diarrhoea; as such, it was not included in the final multivariable model.Figure 2 Risk factors for in-hospital mortality in children with diarrhoea and dehydration (A) Association of each covariable with in-hospital mortality in models using imputed data. (B) Association with in-hospital mortality after adjustment for all patient-level covariables. *Odds ratios not calculated. †Proxy measure for illness severity. In model II, with adjustment of all patient-level covariables, age of 12 months or younger (adjusted-OR [AOR] 1·71, 95% CI 1·42–2·06), female sex (AOR 1·41, 1·19–1·66), diarrhoea duration of more than 14 days (2·10, 1·42–3·12), abnormal respiratory signs (3·62, 2·95–4·44), abnormal circulatory signs (2·29, 1·89–2·77), pallor (2·15, 1·76–2·62), use of intravenous fluids (1·68, 1·41–2·00), and abnormal neurological signs (3·07, 2·54–3·70) remained associated with in-hospital mortality within hospital clusters (figure 2). Notably, within this population, all of whom had diarrhoea and dehydration, individual signs of dehydration and presence of both sunken eyes and delayed skin pinch (severe dehydration) were not independently associated with in-hospital death (severe dehydration AOR 1·08, 0·87–1·35) within hospital clusters. However, when the same model was used in an analysis of the broader population of all patients with diarrhoea or dehydration (n=19 839; figure 1), signs of dehydration were independently associated with in-hospital death (AOR 1·22, 1·01–1·47; further data available on request). Length of illness of more than 2 days, abnormal airway signs, malaria diagnosis, and residence within a malaria endemic zone were not associated with death in the multivariable analysis (model II). Similar associations were seen when the multivariable model was done with the outcome as early in-hospital death, with the exception of diarrhoea duration of more than 14 days, which no longer showed a clear association with mortality (AOR 1·43, 0·87–2·34). When bloody diarrhoea was added to the final multivariable model, it was not associated with the outcome and the effect of the other covariates was not altered.

hospital death, with the exception of diarrhoea duration of more than 14 days, which no longer showed a clear association with mortality (AOR 1·43, 0·87–2·34). When bloody diarrhoea was added to the final multivariable model, it was not associated with the outcome and the effect of the other covariates was not altered. In analyses restricted to the stratum for whom the fluid prescription was correct, the strength of association (ORs) with early in-hospital mortality was, in general, considerably lower than in the stratum in which fluid prescriptions were incorrect, even after adjustment for use of intravenous fluids (a proxy for severity). In fact, positive multiplicative interactions (OR>1) for association of incorrect fluid management with death in children with dehydration signs (OR 1·50, 95% CI 0·79–2·88), respiratory signs (OR 1·23, 0·68–2·24), and pallor (OR 1·70, 0·95–3·02) are evidence of interaction (figure 3). Furthermore, likely presence of additive interaction (relative excess odds due to interaction is not zero) point to the potential public health benefit of being able to correctly prescribe fluid regimens (table 2). The predicted probability of early in-hospital death was reduced by 6% when derived from model II with correct fluid prescription compared with when fluid prescription is wrong.Figure 3 Risk factors and interaction with fluid management for early in-hospital mortality Table 2 Interactions of fluid management with risk factors for in-hospital early deaths

In analyses restricted to the stratum for whom the fluid prescription was correct, the strength of association (ORs) with early in-hospital mortality was, in general, considerably lower than in the stratum in which fluid prescriptions were incorrect, even after adjustment for use of intravenous fluids (a proxy for severity). In fact, positive multiplicative interactions (OR>1) for association of incorrect fluid management with death in children with dehydration signs (OR 1·50, 95% CI 0·79–2·88), respiratory signs (OR 1·23, 0·68–2·24), and pallor (OR 1·70, 0·95–3·02) are evidence of interaction (figure 3). Furthermore, likely presence of additive interaction (relative excess odds due to interaction is not zero) point to the potential public health benefit of being able to correctly prescribe fluid regimens (table 2). The predicted probability of early in-hospital death was reduced by 6% when derived from model II with correct fluid prescription compared with when fluid prescription is wrong.Figure 3 Risk factors and interaction with fluid management for early in-hospital mortality Table 2 Interactions of fluid management with risk factors for in-hospital early deaths Multiplicative interaction Fluid wrong and symptom present (OR10) Fluid right and symptom present (OR11) Fluid right and symptom absent (OR01) RERI Attributable proportion Pallor 1·70 (0·95 to 3·02) 2·16 (1·66 to 2·82) 1·07 (0·69 to 1·67) 0·29 (0·21 to 0·41) −0·38 (−1·07 to 0·31) −0·35 (−1·09 to 0·38) Circulatory 0·95 (0·53 to 1·73) 1·92 (1·47 to 2·51) 0·64 (0·40 to 1·03) 0·35 (0·25 to 0·50) −0·63 (−1·19 to −0·06) −0·98 (−2·06 to 0·11) Dehydration signs only 1·50 (0·79 to 2·88) 0·96 (0·70 to 1·33) 0·38 (0·26 to 0·55) 0·26 (0·15 to 0·45) 0·25 (−0·18 to 0·48) 0·40 (−0·53 to 1·33) Neurological 0·86 (0·51 to 1·48) 3·57 (2·75 to 4·64) 1·15 (0·77 to 1·72) 0·37 (0·25 to 0·54) −1·80 (−2·71 to −0·88) −1·56 (−2·61 to −0·52) Respiratory 1·23 (0·68 to 2·24) 3·18 (2·35 to 4·31) 1·17 (0·80 to 1·73) 0·30 (0·18 to 0·49) −1·31 (−2·16 to −9·45) −1·11 (−1·90 to −0·33) Interaction exists if RERI is not zero (RERI=OR11 – OR10 – OR01 + 1). A negative value of RERI implies reduced risk due to interaction with fluid treatment. The attributable proportion is a measure of the proportion of the risk in the doubly exposed group that is due to the interaction itself. An interaction exists if the attributable proportion is not zero (attributable proportion=RERI/OR11). A negative value of the attributable proportion suggests that interaction reduces the risk of outcome (death). Multiplicative interaction=OR11/(OR10 ×   OR01). Presence of interaction on either additive or multiplicative scales indicates that both exposures have an effect on the outcome. The reference group was fluid wrong and symptom absent (OR00)=1·00. RERI=relative excess risk due to interaction.

the risk of outcome (death). Multiplicative interaction=OR11/(OR10 ×   OR01). Presence of interaction on either additive or multiplicative scales indicates that both exposures have an effect on the outcome. The reference group was fluid wrong and symptom absent (OR00)=1·00. RERI=relative excess risk due to interaction. Discussion A study6 using the Child Health and Nutrition Research Initiative's methodology recently identified risk factors for diarrhoeal deaths as a research priority. This study investigated clinical features associated with mortality and whether, for early deaths, associations are modified by prescribed fluid therapy in accordance with WHO and Kenyan guidance in children with both diarrhoea and dehydration. We used routinely collected data from 13 first-level referral hospitals in Kenya involved in a collaborative effort to improve availability of routine data on admitted children and the management of common conditions.7, 8, 9 We excluded children who had severe acute malnutrition, a known major risk factor for mortality with specific fluid management guidance,7, 8 and children younger than 30 days or older than 5 years because no standard rehydration guidelines exist for those ages. We did not have microbial cultures of stool or blood but malaria testing is done consistently by these hospitals in their own laboratories.9

for mortality with specific fluid management guidance,7, 8 and children younger than 30 days or older than 5 years because no standard rehydration guidelines exist for those ages. We did not have microbial cultures of stool or blood but malaria testing is done consistently by these hospitals in their own laboratories.9 In our study of 8562 children admitted with both diarrhoea and dehydration, we found that most (>80%) had at least one additional diagnosis (comorbidity). We also note that diarrhoea not accompanied by clinically diagnosed dehydration is a common symptom in a range of diseases. Among participants with diarrhoea and dehydration (representing 16% of all admissions and associated with 28% of all death), respiratory signs, circulatory impairment, pallor, neurological signs and symptoms, prolonged diarrhoea, age younger than 1 year, and female sex were associated with increased risk of in-hospital death. Our findings are consistent with studies that have shown impaired consciousness, convulsions,18 signs of pneumonia,19, 20 anaemia, weak pulse volume,20 longer duration of preadmission illness,21 and persistent diarrhoea22, 23, 24, 25 as risk factors for mortality in children with diarrhoea. Respiratory signs might in fact be indicative of pneumonia comorbidity. Female sex and younger age (<1 year), which were significantly associated with mortality in this study, have also been found to be risk factors in other studies. Our study did not investigate reasons for higher mortality in female participants than in male participants; some of the reasons suggested include social inequality and gender bias, but whether these factors are also important in Kenya is unclear.26

this study, have also been found to be risk factors in other studies. Our study did not investigate reasons for higher mortality in female participants than in male participants; some of the reasons suggested include social inequality and gender bias, but whether these factors are also important in Kenya is unclear.26 In our analysis of children with information on fluid prescribing, we found that correct fluid prescription was associated with reduced odds of death in participants with various clinical signs. Recommended fluid regimens are more complex for children who are more severely ill and might more often be wrongly prescribed. As well as adjusting for specific signs of severity, we also adjusted for the choice to use intravenous fluid therapy as a possible additional indication of the perceived disease severity. Overall, our findings suggest a potential public health benefit from ensuring or improving correct fluid prescription practices. In the exploratory analyses (data not shown), most incorrect fluid prescriptions in this study involved inadequate volumes or use of recommended volumes that were given over longer than recommended durations. Any efforts to improve correct fluid management will have to deal with the problem of comorbidity in children with diarrhoea and dehydration, something not covered in existing guidelines. The importance of tailoring fluid therapy to specific populations has been highlighted by the FEAST study. This large multicentre clinical trial of fluid treatment in African children with impaired perfusion and febrile non-diarrhoeal illnesses showed rapid fluid administration to be harmful in non-diarrhoeal cases.27 Special precaution was unlikely in the data presented because the data presented here represent prescription by front-line clinicians (mainly intern doctors or non-physician clinicians) who generally rely on guidelines and any decisions to deviate are made by senior clinicians much later after admission and are not reflected in this analysis.

nlikely in the data presented because the data presented here represent prescription by front-line clinicians (mainly intern doctors or non-physician clinicians) who generally rely on guidelines and any decisions to deviate are made by senior clinicians much later after admission and are not reflected in this analysis. Case fatality in the present study (9%) is similar to that found in a study28 involving two hospitals in western Kenya (9%); however, the patient population in the other study was different from ours because it included all children with diarrhoea and children with malnutrition. The study28 in western Kenya and the recent GEMS studies have shown that bacterial pathogens, including non-typhoidal Salmonella, Cryptosporidium, Shigella, and E coli, are important risk factors for death and severe disease and that rotavirus was not an important cause of in-hospital mortality.5 These findings have led to calls for studies on bacterial aetiology of diarrhoea and studies on antimicrobial resistance. Like the study in western Kenya, in our study, antibiotics were given to 60% of children with diarrhoea or dehydration, with comorbidities providing an indication for antibiotic use; however, most of the antibiotics used might not be effective against gut infections. Antibiotics are not routinely recommended and their use in the present study was a proxy for comorbidity.

otics were given to 60% of children with diarrhoea or dehydration, with comorbidities providing an indication for antibiotic use; however, most of the antibiotics used might not be effective against gut infections. Antibiotics are not routinely recommended and their use in the present study was a proxy for comorbidity. The CIN has enabled improvements in documentation of the assessment, diagnosis, and treatment at admission of conditions since its inception and it uses a robust data capture system to try and optimise data quality.7, 8, 9 However, some data are missing and we employed imputation to maximise use of all available data. We do not have insight into whether the treatments prescribed at admission are accurately given—something that can be compromised by resource unavailability and staff shortages.29, 30 We also cannot be sure that clinicians are correct in their diagnosis and assessment or that they prescribe the right fluid regimens to the right patients. Data from such routine settings do not provide any insight on microbial aetiological diagnoses or on biochemical tests that might help assess severity of disease. Multiple imputation for missing data is based on an assumption that data are missing at random, which is difficult to ascertain. Finally, CIN does not collect data on all the care children receive across the period of their admission. Despite these limitations, this large dataset does offer a picture of routine admission characteristics and treatments across multiple sites and over a prolonged period of time, and findings might therefore be generalisable to similar African settings.

all the care children receive across the period of their admission. Despite these limitations, this large dataset does offer a picture of routine admission characteristics and treatments across multiple sites and over a prolonged period of time, and findings might therefore be generalisable to similar African settings. Children with diarrhoea and dehydration who are at most risk of in-hospital death are those with other signs of severe disease (comorbities) as opposed to those with uncomplicated dehydration. Importantly, prescription of recommended rehydration guidelines reduces risk of death in this group of children. However, correct fluid prescription might be an indicator of improved attention to the provision of good overall care, including other treatments such as antibiotics and nursing care, rather than fluid management alone. Strategies to optimise delivery of recommended guidance should be accompanied by studies on the management of dehydration in children with comorbidities, the vulnerability of female children, and the delivery of immediate care. Future research that investigates diarrhoea aetiology, that includes an enhanced range of diagnostic tests (including for comorbidities), and that effectively standardises the delivery of fluid therapy in accordance with guidelines, would help tease out more specific risk factors for poor outcome and could support newer management approaches that target more specific patient groups.

des an enhanced range of diagnostic tests (including for comorbidities), and that effectively standardises the delivery of fluid therapy in accordance with guidelines, would help tease out more specific risk factors for poor outcome and could support newer management approaches that target more specific patient groups. Acknowledgments We would like to thank the Ministry of Health who gave permission for this study to be developed and have supported the implementation of the Clinical Information Network together with the county health executives and all hospital management teams. We are grateful to the Kenya Paediatric Association, the Kenya Ministry of Health, and the University of Nairobi for promoting the aims of the Clinical Information Network and the support they provided through their officers and membership. We also thank the hospital clinical teams on all the paediatric wards who provide care to the children for whom this project is designed. This study is published with the permission of the director of Kenya Medical Research Institute (KEMRI). Funds from The Wellcome Trust (#097170) awarded to ME as a Senior Fellowship together with additional funds from a Wellcome Trust core grant awarded to the KEMRI-Wellcome Trust Research Programme (#092654) supported this study. SA is supported by the Initiative to Develop African Research Leaders (IDeAL) Wellcome Trust award (#107769/Z/15/Z).

Trust (#097170) awarded to ME as a Senior Fellowship together with additional funds from a Wellcome Trust core grant awarded to the KEMRI-Wellcome Trust Research Programme (#092654) supported this study. SA is supported by the Initiative to Develop African Research Leaders (IDeAL) Wellcome Trust award (#107769/Z/15/Z). Contributors PA, DG, AA, GI, KS, and ME made suggestions on appropriate analytic approach and helped interpret the results, improved the initial drafts of the manuscript, contributed to the manuscript's development, and approved the final version. SA conceived the study, did the analyses with support of DG, AA, GI, KS and ME, and drafted the initial manuscript. The Clinical Information Network authors contributed to the design of the data collection tools, conduct of the work, collection of data, and data quality assurance that form the basis of this report, and saw and approved the report's findings. All coauthors reviewed and approved the final version of the manuscript.

t. The Clinical Information Network authors contributed to the design of the data collection tools, conduct of the work, collection of data, and data quality assurance that form the basis of this report, and saw and approved the report's findings. All coauthors reviewed and approved the final version of the manuscript. Clinical Information Network authors The Clinical Information Network team who contributed to this study as part of the Clinical Information Network author group include Samuel Ngarngar (Vihiga County Hospital); Nick Aduro (Kakamega County Hospital); Loice Mutai and David Kimutai (Mbagathi County Hospital); Caren Emadau, Cecilia Mutiso, and Celia Muturi (Mama Lucy Kibaki County Hospital); Charles Nzioki (Machakos County Hospital); Francis Kanyingi and Agnes Mithamo (Nyeri County Hospital); Magdalene Kuria (Kisumu East County Hospital); Samuel Otido and Anne Kamunya (Embu County Hospital); Alice Kariuki (Karatina County Hospital); Peris Njiiri (Kerugoya County Hospital); Rachel Inginia and Melab Musabi (Kitale County Hospital); Barnabas Kigen (Busia County Hospital); Grace Akech Ochieng and Lydia Thuranira (Kiambu County Hospital); Morris Ogero; Thomas Julius; Boniface Makone; Mercy Chepkirui; and James Wafula (Kenya Medical Research Institute-Wellcome Trust Research Programme). Declaration of interests We declare no competing interests.

Clinical Information Network authors The Clinical Information Network team who contributed to this study as part of the Clinical Information Network author group include Samuel Ngarngar (Vihiga County Hospital); Nick Aduro (Kakamega County Hospital); Loice Mutai and David Kimutai (Mbagathi County Hospital); Caren Emadau, Cecilia Mutiso, and Celia Muturi (Mama Lucy Kibaki County Hospital); Charles Nzioki (Machakos County Hospital); Francis Kanyingi and Agnes Mithamo (Nyeri County Hospital); Magdalene Kuria (Kisumu East County Hospital); Samuel Otido and Anne Kamunya (Embu County Hospital); Alice Kariuki (Karatina County Hospital); Peris Njiiri (Kerugoya County Hospital); Rachel Inginia and Melab Musabi (Kitale County Hospital); Barnabas Kigen (Busia County Hospital); Grace Akech Ochieng and Lydia Thuranira (Kiambu County Hospital); Morris Ogero; Thomas Julius; Boniface Makone; Mercy Chepkirui; and James Wafula (Kenya Medical Research Institute-Wellcome Trust Research Programme). Declaration of interests We declare no competing interests. * Abnormal airway signs include obstructed (stridor) or absent breathing, central cyanosis, or severe respiratory distress. Only stridor was included in the analysis as the other signs were rare while severe respiratory distress is not collected in the database.

Introduction Psychotic experiences, such as delusions or hallucinations (core symptoms of schizophrenia), are common in the general population, especially in childhood and adolescence.1 Findings from a meta-analysis1 suggested that the median prevalence of psychotic experiences in late childhood and early adolescence (age 9–12 years) is 17%, decreasing to 7·5% in late adolescence1 and reaching a prevalence of around 5% in adulthood.2 Traditionally, psychotic experiences have been predominantly conceptualised as prodromal expressions of psychotic disorders. However, an increasing number of studies have shown that they are also longitudinally associated with a range of adverse psychopathological outcomes—including mood disorders,3 substance use,3 post-traumatic stress disorder,5 and suicide attempts5, 6—suggesting that they could represent early transdiagnostic, non-specific markers of mental health disorders.7 Research in context Evidence before this study

Introduction Psychotic experiences, such as delusions or hallucinations (core symptoms of schizophrenia), are common in the general population, especially in childhood and adolescence.1 Findings from a meta-analysis1 suggested that the median prevalence of psychotic experiences in late childhood and early adolescence (age 9–12 years) is 17%, decreasing to 7·5% in late adolescence1 and reaching a prevalence of around 5% in adulthood.2 Traditionally, psychotic experiences have been predominantly conceptualised as prodromal expressions of psychotic disorders. However, an increasing number of studies have shown that they are also longitudinally associated with a range of adverse psychopathological outcomes—including mood disorders,3 substance use,3 post-traumatic stress disorder,5 and suicide attempts5, 6—suggesting that they could represent early transdiagnostic, non-specific markers of mental health disorders.7 Research in context Evidence before this study We searched MEDLINE, PubMed, and Google Scholar for articles published between 2003–18, on the comorbidity of psychotic symptoms (search terms “psychotic symptoms” and “psychosis”) and eating disorders (“eating disorders”, “anorexia”, “bulimia”, and “binge eating”). We noted three studies that had investigated this association. These studies suggested that both adolescents and adults with psychotic symptoms were more likely to experience eating-disorder behaviours (and vice-versa) or have a comorbid diagnosis of an eating disorder over the course of their lifetime. One study showed that adolescents and adults with psychotic symptoms were more likely to develop bulimia nervosa. To date, studies have predominantly investigated this association cross-sectionally, or relied on study participant recall of onset of symptoms. From these approaches it is difficult to establish temporality of association—a precondition for causal inference—and exclude recall bias. Finally, most existing studies investigated eating-disorder diagnoses, but not eating disorder behaviours, which might be more common in the general population.

f onset of symptoms. From these approaches it is difficult to establish temporality of association—a precondition for causal inference—and exclude recall bias. Finally, most existing studies investigated eating-disorder diagnoses, but not eating disorder behaviours, which might be more common in the general population. Added value of this study To our knowledge, ours is the first study to investigate the association of psychotic experiences and disordered-eating behaviours longitudinally in a large sample of English adolescents. We noted evidence that children with psychotic experiences in early adolescence had increased odds of reporting disordered eating in late adolescence and having more severe presentations, denoted by a greater number of disordered-eating behaviours. Compared with previous studies, our approach has the advantage of being able to exclude recall bias, limit the potential for reverse causation, and control for several potential child and familial confounding factors. Implications of all the available evidence

To our knowledge, ours is the first study to investigate the association of psychotic experiences and disordered-eating behaviours longitudinally in a large sample of English adolescents. We noted evidence that children with psychotic experiences in early adolescence had increased odds of reporting disordered eating in late adolescence and having more severe presentations, denoted by a greater number of disordered-eating behaviours. Compared with previous studies, our approach has the advantage of being able to exclude recall bias, limit the potential for reverse causation, and control for several potential child and familial confounding factors. Implications of all the available evidence Despite little research attention, evidence suggests that psychotic symptoms and eating disorders can co-occur, as observed in clinical reports and highlighted in the epidemiological literature. Psychotic experiences in adolescence have been previously shown to be a non-specific early marker of adverse psychopathologies, which we extend to include eating disorders. Future studies should investigate the presence of common and specific risk factors which may underpin their causes. The longitudinal associations we observed between psychotic experiences and disordered eating could indicate the presence of shared aetiological mechanisms. Furthermore, the greater prevalence of bingeing, fasting, and purging behaviours among children with psychotic experiences could partly explain the higher rates of metabolic disorders in individuals with psychotic illnesses or indicate shared metabolic risk factors, although further research in this area is warranted.

Furthermore, the greater prevalence of bingeing, fasting, and purging behaviours among children with psychotic experiences could partly explain the higher rates of metabolic disorders in individuals with psychotic illnesses or indicate shared metabolic risk factors, although further research in this area is warranted. Eating disorders (anorexia nervosa, bulimia nervosa, and binge-eating disorder) are comorbid with mood disorders, substance use, post-traumatic stress disorder, and suicide attempts;8, 9, 10 nevertheless, their association with psychotic experiences has received little attention in the epidemiological literature. To our knowledge, only two studies have investigated this association in the general population, both of which were based on adult samples.4, 11 One cross-sectional survey showed that participants who screened positive for eating disorders using the SCOFF five-item screening questionnaire for eating disorders had increased odds of reporting psychotic experiences and hypomania, and that those reporting loss of control when eating (which is a symptom of binge eating) had greater odds of experiencing auditory hallucinations and other psychotic experiences.4 In line with this finding, results from a multisite study showed that psychotic symptoms were associated with twice the odds of subsequently experiencing bulimia nervosa, an eating disorder characterised by recurrent episodes of binge eating and purging (ie, self-induced vomiting).11 Finally, findings from another study showed that adolescents with clinical diagnoses of anorexia and bulimia nervosa were more likely to have psychotic experiences than general population controls.12 These findings are largely similar to those from clinical studies showing a comorbidity between psychotic illnesses, including schizophrenia and bipolar disorder, and disordered eating behaviours and diagnoses of eating disorders.13, 14

were more likely to have psychotic experiences than general population controls.12 These findings are largely similar to those from clinical studies showing a comorbidity between psychotic illnesses, including schizophrenia and bipolar disorder, and disordered eating behaviours and diagnoses of eating disorders.13, 14 Although evidence suggests that psychotic experiences and eating disorders can co-occur,3, 11, 12 several methodological limitations of these studies preclude causal inferences being made from these associations. For instance, the cross-sectional3, 12 and retrospective designs11 of these studies prevented the investigation of temporal associations. Additionally, because of the under-representation of eating disorders in clinical settings, the use of clinical samples12 could also have resulted in selection bias and overlooked psychopathological presentations more commonly observed in the general population. Longitudinal research on the association between psychotic experiences and disordered eating is therefore needed to elucidate the temporality of these associations and their specificity to individual disordered-eating behaviours, which are both necessary preconditions for causal inference. We investigated whether psychotic experiences at age 13 years were prospectively associated with disordered-eating behaviours at age 18 years in a longitudinal, general population sample of UK adolescents. Based on previous evidence we hypothesised that psychotic experiences would be associated with greater disordered eating behaviours and body-mass index (BMI).

s at age 13 years were prospectively associated with disordered-eating behaviours at age 18 years in a longitudinal, general population sample of UK adolescents. Based on previous evidence we hypothesised that psychotic experiences would be associated with greater disordered eating behaviours and body-mass index (BMI). Methods Data sample We used data from the Avon Longitudinal Study of Parents and Children (ALSPAC), a longitudinal birth cohort based in Avon (England, UK). All mothers in this region with an expected delivery date between April 1, 1991, and Dec 31, 1992 were eligible and invited to participate in the study. 14 541 pregnant women were recruited for the study and resulted in 13 988 live births at 1 year. Information on these children and their mothers' health and life circumstances has since been collected through clinical visits and self-reported questionnaires. More details on the ALSPAC sample and measures collected have been previously published.15 The local research ethics committees at the University of Bristol (Bristol, England, UK) and the ALSPAC Law and Ethics Committee provided ethical approval for this study. Parents provided written consent for their children's participation in the study. The ALSPAC study website contains details of all the data that are available through a fully searchable dictionary on the study website.

istol, England, UK) and the ALSPAC Law and Ethics Committee provided ethical approval for this study. Parents provided written consent for their children's participation in the study. The ALSPAC study website contains details of all the data that are available through a fully searchable dictionary on the study website. Exposure Data on psychotic experiences were collected at clinic assessments done when children were nearly 13 years old using the psychotic-like symptoms interview, a semi-structured interview consisting of 12 questions adapted from the Diagnostic Interview Schedule for Children16 and the Schedules for Clinical Assessment in Neuropsychiatry17 assessing the presence of hallucinations, delusions, and thought interference. From the total interview score, a binary variable was created to indicate the absence of psychotic experiences or the presence of suspected or definite psychotic symptoms. Children were defined as having any suspected or definite psychotic experiences if they reported one or more such experiences that could not be attributable to an absence of sleep or fever. This measure has been previously validated18 and used extensively in general population studies,19, 20, 21, 22 thus ensuring comparability of our findings.

having any suspected or definite psychotic experiences if they reported one or more such experiences that could not be attributable to an absence of sleep or fever. This measure has been previously validated18 and used extensively in general population studies,19, 20, 21, 22 thus ensuring comparability of our findings. Outcomes Information on disordered-eating behaviours was collected via postal questionnaires at age approximately 18 years using adapted questions of the Youth Risk Behaviour Surveillance System questionnaire.23 We derived four binary variables indicating the presence or absence of binge eating, purging, excessive exercise, and fasting in the previous 12 months. Binge eating was defined as having ever eaten large amounts of food in short periods of time with usually a sense of loss of control during these episodes. Children were considered as purging if they had ever had self-induced vomiting or had used laxatives for weight loss, and as fasting if they had ever fasted for a whole day to lose weight. Excessive exercise was defined as exercising frequently for weight loss even when sick, or feeling guilty when not exercising, as these dimensions (as opposed to objective amounts of exercise undertaken, for example) have been shown to underlie eating-disorder psychopathology.24 A more detailed description of the original questions included in the age 18 years questionnaire and how we derived our outcome variables is in the appendix. Because our individual outcomes were rare, we also investigated the presence of any disordered eating behaviours to increase the statistical power, and the number of disordered eating behaviours (from none to four) to explore the presence of an association between psychotic experiences and more severe presentations of eating disorders, defined as a greater number of disordered-eating behaviours. A continuous standardised measure of BMI was derived from objective measures of height and weight obtained at clinic visits that took place when the child was approximately age 18 years.

psychotic experiences and more severe presentations of eating disorders, defined as a greater number of disordered-eating behaviours. A continuous standardised measure of BMI was derived from objective measures of height and weight obtained at clinic visits that took place when the child was approximately age 18 years. Confounders Analyses were adjusted for several child and maternal characteristics that we hypothesised could confound the association between psychotic experiences and eating-disorder behaviours, based on findings from previous studies (appendix). We modelled our assumptions using a direct acyclical graph25 and the web programme DAGgitty (appendix).26 We included child's sex (male or female), autistic traits at age 7 years, and BMI and depressive symptoms at age 13 years. To measure autistic traits in the children, we used the Social and Communication Disorder Checklist (SCDC), a parent-reported 12-item scale that measures deficits in verbal and non-verbal social communication, typically observed in autism spectrum disorders.27 Total SCDC scores ranged from 0 to 24, with higher scores denoting greater social communication difficulties. Depressive symptoms were measured using the short Moods and Feelings Questionnaire, a self-reported 13-item questionnaire with scores ranging from 0 to 26, with greater scores indicating more severe depressive symptoms.28 We further adjusted for the maternal characteristics of age, marital status (single, married, divorced, separated, or widowed), and highest academic qualification (up to secondary school, or university degree or higher), all measured at birth of the child. We also included a continuous measure of maternal depressive symptoms at 32 weeks of gestation using the Edinburgh Postnatal Depression Scale, a ten-item questionnaire from which we derived a continuous measure ranging from 0 to 30; this scale has been previously validated and extensively used in epidemiological research.29 Finally, we adjusted for baseline BMI at age 13 years, which was measured using standardised age and gender-specific Z scores using the updated Stata package zanthro,30 as recommended in previous studies.31, 32

ng from 0 to 30; this scale has been previously validated and extensively used in epidemiological research.29 Finally, we adjusted for baseline BMI at age 13 years, which was measured using standardised age and gender-specific Z scores using the updated Stata package zanthro,30 as recommended in previous studies.31, 32 Statistical analysis We used frequencies (with percentages), means (with SDs), and medians (with IQRs) to describe the distribution of confounders and outcomes in relation to the exposure. To investigate the association between psychotic experiences and disordered eating behaviours (ie, binge eating, purging, fasting, excessive exercise, with any behaviours), we used a univariable (crude model) and four multivariable logistic regressions models progressively adjusting for maternal and child confounders. In adjusted model 1, we controlled for children's sex, and maternal characteristics of age, social class, marital status, and depressive symptoms. In adjusted model 2, we further controlled for autistic traits; in adjusted model 3 we subsequently included BMI at age 13 years; and in adjusted model 4, child's depressive symptoms. To model the associations between psychotic experiences and BMI we used linear regressions, whereas for the number of disordered-eating behaviours we used negative binomial regression models, as these were best suited to model count data in the presence of overdispersion of data (ie, when the SD is greater than the mean), following the same model-building approach described above.

we used linear regressions, whereas for the number of disordered-eating behaviours we used negative binomial regression models, as these were best suited to model count data in the presence of overdispersion of data (ie, when the SD is greater than the mean), following the same model-building approach described above. We explored the effect of missing data using multiple imputation by chained equations.33 We imputed missing covariate and outcome data using linear, logistic, and multinomial logistic regression models (as appropriate) using the MI impute chained Stata command and 50 imputed dataset, using Rubin's rules34 to estimate combined effect sizes. In our imputation models we included all variables used in our main model and the number of auxiliary variables putatively associated with the imputed variables and outcome missingness, as recommended in a previous study.35 These variables were low birthweight (<2500 g or ≥2500 g); intelligence quotient at age 7 years; BMI and disordered behaviours at age 16 years; a continuous measure of depressive symptoms, measured with the short Moods and Feelings Questionnaire at age 16 and 18 years;36, 37 polygenic risk scores for schizophrenia;38 a continuous measure of negative symptoms at age 16 years measured with ten items from the Community Assessment of Psychic Experiences; and sexual orientation as this has been shown to be associated with disordered eating in this sample39 as well as missingness.

8 years;36, 37 polygenic risk scores for schizophrenia;38 a continuous measure of negative symptoms at age 16 years measured with ten items from the Community Assessment of Psychic Experiences; and sexual orientation as this has been shown to be associated with disordered eating in this sample39 as well as missingness. Because the results of the imputed models did not differ qualitatively from those of complete case analyses (ie, point estimates were similar and 95% CI overlapped), in the results section we present models based on a sample of participants with complete exposure data and imputed confounders and outcome (sample A) and discuss any observed differences in results compared with complete cases analyses (sample B). We also compared these results with a set of analyses done on a sample of children with complete exposure and outcome data, but also missing, and hence imputed, confounders (sample C). All analyses were done using Stata v.13. Role of the funding source The funder of the study had no role in study design, data collection, data analysis, data interpretation, or writing of the report. The corresponding author had full access to all the data in the study and had final responsibility for the decision to submit for publication.

Because the results of the imputed models did not differ qualitatively from those of complete case analyses (ie, point estimates were similar and 95% CI overlapped), in the results section we present models based on a sample of participants with complete exposure data and imputed confounders and outcome (sample A) and discuss any observed differences in results compared with complete cases analyses (sample B). We also compared these results with a set of analyses done on a sample of children with complete exposure and outcome data, but also missing, and hence imputed, confounders (sample C). All analyses were done using Stata v.13. Role of the funding source The funder of the study had no role in study design, data collection, data analysis, data interpretation, or writing of the report. The corresponding author had full access to all the data in the study and had final responsibility for the decision to submit for publication. Results We obtained data on psychotic experiences at age 13 years for 6361 children (table 1). Of these, most were girls and had mothers educated up to secondary school who were married and had non-manual occupation. A greater proportion of children with psychotic experiences at 13 years were girls, reported greater depressive symptoms and autistic traits, had younger mothers who were single, separated, or widowed, and had lower levels of education and higher depressive symptoms.Table 1 Descriptive statistics of psychotic experiences at age 13 years with confounding and outcome variables

es at 13 years were girls, reported greater depressive symptoms and autistic traits, had younger mothers who were single, separated, or widowed, and had lower levels of education and higher depressive symptoms.Table 1 Descriptive statistics of psychotic experiences at age 13 years with confounding and outcome variables Total* Psychotic experiences at 13 years No Yes Total 6361 (100%) 5627 (89%) 734 (12%) Sex Male 3131 (49%) 2794 (50%) 337 (46%) Female 3230 (51%) 2833 (50%) 397(54%) Maternal marital status Single (never married) 863 (14%) 733 (13%) 130 (19%) Married 5046 (81%) 4523 (82%) 523 (74%) Widowed, divorced, or separated 302 (5%) 251 (5%) 51 (7%) Maternal highest academic qualification Compulsory education 5133 (84%) 4518 (83%) 615 (87%) Degree or higher 1005 (16%) 914 (17%) 91 (13%) Binge eating at 18 years No 2474 (96%) 2236 (96%) 238 (92%) Yes 107 (4%) 87 (4%) 20 (8%) Purging at 18 years No 2408 (94%) 2177 (94%) 231 (90%) Yes 156 (6%) 130 (6%) 26 (10%) Fasting at 18 years No 2297 (89%) 2089 (90%) 208 (80%) Yes 291 (11%) 240 (10%) 51 (20%) Excessive exercise at 18 years No 2472 (96%) 2220 (96%) 252 (97%) Yes 101 (4%) 93 (4%) 8 (3%) Any disordered eating at 18 years No 2079 (82%) 1899 (83%) 180 (71%) Yes 453 (18%) 380 (17%) 73 (29%) Maternal age 29·1 (4·6) 29·2 (4·5) 28·5 (4·8) Maternal depressive symptoms 6·4 (4·6) 6·3 (4·5) 7·4 (4·8) Child's BMI (Z scores) At 13 years 0·3 (1·1) 0·3 (1·1) 0·3 (1·1) At 18 years –0·3 (1·0) –0·3 (1·0) –0·2 (0·9) Child's depressive symptoms at age 13 years 3 (1–5) 3 (1–5) 6 (3–10) Child's autistic traits at age 7 years 2 (0–4) 1 (0–4) 2 (0–4) Number of disordered-eating behaviours 6 (3–9) 6 (3–9) 7 (4–11) Data are n (%), mean (SD), or median (IQR). BMI=body-mass index.

1·1) 0·3 (1·1) At 18 years –0·3 (1·0) –0·3 (1·0) –0·2 (0·9) Child's depressive symptoms at age 13 years 3 (1–5) 3 (1–5) 6 (3–10) Child's autistic traits at age 7 years 2 (0–4) 1 (0–4) 2 (0–4) Number of disordered-eating behaviours 6 (3–9) 6 (3–9) 7 (4–11) Data are n (%), mean (SD), or median (IQR). BMI=body-mass index. * Distribution (ie, numbers and proportions) of confounding and outcome variables refers to the sample of children with available exposure data (our analytical sample) shown in the first row of the table. Among children with complete exposure data, around 30% did not have any outcome data at age 18 years (appendix). Children who were male, with greater autistic traits at age 7 years (and who had a mother who had only completed secondary education, was single, younger, and had greater depressive symptoms), had greater odds of not having outcome data available at age 18 years. With the exception of excessive exercise all disordered eating behaviours were more prevalent among children with psychotic experiences at age 13 years. Overall, around a third of children with psychotic experiences reported any disordered eating behaviours, compared with 17% of children without psychotic experiences (table 1). Children with psychotic experiences also reported more disordered eating behaviours than those without psychotic experiences (median 7 [IQR 4–11] vs 6 [IQR 3–9]).

hird of children with psychotic experiences reported any disordered eating behaviours, compared with 17% of children without psychotic experiences (table 1). Children with psychotic experiences also reported more disordered eating behaviours than those without psychotic experiences (median 7 [IQR 4–11] vs 6 [IQR 3–9]). Children who reported psychotic experiences at 13 years had an increased risk of having any disordered eating behaviours at age 18 years in crude models (table 2) and in models adjusting for child and maternal confounders, autistic traits, BMI at 13 years, and depressive symptoms (table 2). Psychotic experiences at age 13 years were also associated with greater number of disordered eating behaviours in the crude model and in models adjusting for child and maternal confounders including depressive symptoms at baseline (ie, at age 13 years when psychotic experiences were measured).Table 2 Logistic, linear (for BMI), and negative binomial (for number of disordered eating behaviours) regression models

ating behaviours in the crude model and in models adjusting for child and maternal confounders including depressive symptoms at baseline (ie, at age 13 years when psychotic experiences were measured).Table 2 Logistic, linear (for BMI), and negative binomial (for number of disordered eating behaviours) regression models Complete exposure, imputed confounders and outcome (sample A) Odds ratio (95%CI) Complete cases (sample B) Odds ratio (95%CI) Binge eating n=6361 n=1940 Crude model 2·23 (1·36–3·69), p=0·0018 1·79 (0·96–3·30), p=0·062 Adjusted model 1* 2·01 (1·20–3.37), p=0·0082 1·57 (0·84–2·95), p=0·16 Adjusted model 2† 1·99 (1·18–3.33), p=0.0097 1·57 (0·83–2·94), p=0·16 Adjusted model 3‡ 2·04 (1·21–3·43), p=0·0081 1·60 (0·85–3·01), p=0·14 Adjusted model 4§ 1·49 (0·88–2·54), p=0·14 1·27 (0·66–2·45), p=0·48 Purging n=6361 n=1926 Crude model 1·88 (1·24–2·84), p=0·0031 1·99 (1·18–3·35), p=0·0089 Adjusted model 1* 1·74 (1·13–2·68), p=0·013 1·85 (1·08–3·15), p=0·025 Adjusted model 2† 1·73 (1·12–2·67), p=0·013 1·85 (1·08–3·15), p=0·025 Adjusted model 3‡ 1·76 (1·14–2·72), p=0·011 1·85 (1·08–3·16), p=0·025 Adjusted model 4§ 1·51 (0·96–2·36), p=0·071 1·85 (1·05–3·23), p=0·032 Excessive exercise n=6361 n=1929 Crude model 0·84 (0·40–1·78), p=0·65 0·53 (0·19–1·47), p=0·22 Adjusted model 1* 0·79 (0·37–1·71), p=0·55 0·50 (0·18–1·39), p=0·18 Adjusted model 2† 0·79 (0·37–1·71), p=0·55 0·49 (0·18–1·38), p=0·18 Adjusted model 3‡ 0·80 (0·37–1·73), p=0·57 0·49 (0·18–1·39), p=0·18 Adjusted model 4§ 0·71 (0·32–1·57), p=0·39 0·46 (0·16–1·32), p=0·15 Fasting n=6361 n=1943 Crude model 2·25 (1·64–3·09), p<0·0001 2·33 (1·58–3·44), p<0·0001 Adjusted model 1* 2·09 (1·49–2·94), p<0·0001 2·26 (1·51–3·40), p<0·0001 Adjusted model 2† 2·05 (1·46–2·89), p<0·0001 2·25 (1·50–3·38), p<0·0001 Adjusted model 3‡ 2·10 (1·49–2·96), p<0·0001 2·25 (1·50–3·39), p<0·0001 Adjusted model 4§ 1·65 (1·15–2·38), p<0·0001 1·94 (1·26–2·97), p=0·0025 Any disordered eating behaviour n=6361 n=1898 Crude model 1·92 (1·46–2·52), p<0·0001 2·06 (1·41–2·90), p<0·0001 Adjusted model 1* 1·82 (1·35–2·44), p<0·0001 1·97 (1·37–2·83), p<0·0001 Adjusted model 2† 1·80 (1·34–2·41), p<0·0001 1·97 (1·37–2·83), p<0·0001 Adjusted model 3‡ 1·83 (1·36–2·46), p<0·0001 1·98 (1·37–2·87), p<0·0001 Adjusted model 4§ 1·50 (1·10–2·03), p=0·010 1·70 (1·16–2·49), p=0·0070 Number of disordered-eating behaviours¶ n=6361 n=1898 Crude model 0·58 (0·32–0·84), p<0·0001 0·54 (0·19–0·88), p=0·0021 Adjusted model 1* 0·49 (0·23–0·75), p<0·0001 0·46 (0·13–0·79), p=0·0058 Ad

1·83 (1·36–2·46), p<0·0001 1·98 (1·37–2·87), p<0·0001 Adjusted model 4§ 1·50 (1·10–2·03), p=0·010 1·70 (1·16–2·49), p=0·0070 Number of disordered-eating behaviours¶ n=6361 n=1898 Crude model 0·58 (0·32–0·84), p<0·0001 0·54 (0·19–0·88), p=0·0021 Adjusted model 1* 0·49 (0·23–0·75), p<0·0001 0·46 (0·13–0·79), p=0·0058 Ad justed model 2† 0·48 (0·22–0·74), p<0·0001 0·46 (0·14–0·79), p=0·0054 Adjusted model 3‡ 0·49 (0·24–0·75), p<0·0001 0·48 (0·16–0·80), p=0·0036 Adjusted model 4§ 0·32 (0·06–0·57), p=0·017 0·38 (0·05–0·71), p=0·024 BMI¶e n=6361 n=2957 Crude model 0·026 (–0·06–0.11), p=0·54 0·016 (–0·09–0·12), p=0·77 Adjusted model 1* 0·006 (–0·07–0·09), p=0·90 –0·011 (–0·12–0·10), p=0·83 Adjusted model 2† –0·001 (–0·08–0·09), p=0·99 –0·018 (–0·13–0·09), p=0·74 Adjusted model 3‡ –0·003 (–0·06–0.05), p=0·92 0·019 (–0·05–0·09), p=0·58 Adjusted model 4§ 0·001 (–0·06–0·06), p=0·99 0·025 (–0·04–0·09), p=0·47 Models tested the associations between suspected or definite (vs none) psychotic experiences at age 13 years and disordered eating behaviours and body-mass index (BMI) at age 18 years. Data presented are odds ratios (95% CI) from complete case analyses (sample B) and analyses done on a sample of children for whom we had with complete exposure data and imputed confounders and outcomes (sample A, main analyses). * Adjusted model 1: adjusted for maternal: age, marital status, education, social class, depression; and child's sex. † Adjusted model 2: adjusted model 1 plus autistic traits at age 7 years. ‡ Adjusted model 3: adjusted model 2 plus BMI at age 13 years.

Data presented are odds ratios (95% CI) from complete case analyses (sample B) and analyses done on a sample of children for whom we had with complete exposure data and imputed confounders and outcomes (sample A, main analyses). * Adjusted model 1: adjusted for maternal: age, marital status, education, social class, depression; and child's sex. † Adjusted model 2: adjusted model 1 plus autistic traits at age 7 years. ‡ Adjusted model 3: adjusted model 2 plus BMI at age 13 years. § Adjusted model 4: adjusted model 3 plus depressive symptoms at age 13 years. ¶ Data given for this variable are coefficients (95% CI).

* Adjusted model 1: adjusted for maternal: age, marital status, education, social class, depression; and child's sex. † Adjusted model 2: adjusted model 1 plus autistic traits at age 7 years. ‡ Adjusted model 3: adjusted model 2 plus BMI at age 13 years. § Adjusted model 4: adjusted model 3 plus depressive symptoms at age 13 years. ¶ Data given for this variable are coefficients (95% CI). When we looked at individual behaviours, psychotic experiences at 13 years were associated with greater odds of binge eating at 18 years in the crude model (table 2). This association was slightly attenuated by the inclusion of child and maternal confounders; however, after adjustment for baseline depressive symptoms there was no evidence of an association with binge eating. In univariable models, psychotic experiences at 13 years were associated with purging behaviours at age 18 years (table 2). This association decreased slightly after inclusion of child and maternal confounders, and there was no association after inclusion of depressive symptoms at baseline. Psychotic episodes were strongly associated with increased risk of fasting in the univariable model, and multivariable models adjusting for child and maternal confounders and child's depressive symptoms at baseline (table 2). We showed no association between psychotic experiences and excessive exercise or BMI in any of the models.

isodes were strongly associated with increased risk of fasting in the univariable model, and multivariable models adjusting for child and maternal confounders and child's depressive symptoms at baseline (table 2). We showed no association between psychotic experiences and excessive exercise or BMI in any of the models. Between 1898 and 2957 adolescents had complete data on all variables included across different outcomes (table 2), and between 2532 and 4035 adolescents had complete data on exposure and outcome (appendix). For all of our outcomes, the odds ratio and 95% CIs were comparable in magnitude across the three data samples. Complete case analyses seemed to overestimate slightly the association between psychotic experiences and purging behaviours and any disordered eating. Outcomes which were more prevalent (ie, fasting and any disordered eating) yielded strong associations with psychotic experiences at 13 years, despite comparable effect sizes with individual behaviours such as bingeing and purging showing no or weak associations with the exposure. Discussion At age 18 years, a third of adolescents who had psychotic experiences at age 13 years reported some disordered eating behaviours. Compared with those who had not had psychotic experiences, children with psychotic experiences at age 13 years had 1·5 times the risk of reporting any disordered eating behaviours in late adolescence. They were also more likely to report more disordered-eating behaviours.

ars reported some disordered eating behaviours. Compared with those who had not had psychotic experiences, children with psychotic experiences at age 13 years had 1·5 times the risk of reporting any disordered eating behaviours in late adolescence. They were also more likely to report more disordered-eating behaviours. In line with our hypotheses, we noted that psychotic experiences at age 13 years were prospectively associated with binge eating, fasting, and purging. Some associations were weaker, or absent in the final model, possibly because of low statistical power. We did not find evidence of an association between psychotic experiences and greater excessive exercise and BMI at age 18 years. The slight overestimation of the association between psychotic experiences and purging behaviours and any disordered eating from complete case analyses could have been due to sampling biases or small sample size, resulting in greater uncertainty around study estimates. The small sample size could also have affected the strength of some associations.

e association between psychotic experiences and purging behaviours and any disordered eating from complete case analyses could have been due to sampling biases or small sample size, resulting in greater uncertainty around study estimates. The small sample size could also have affected the strength of some associations. The use of a large and well characterised general population cohort of UK adolescents brought several strengths to this study. First, data were collected prospectively using well validated questions, which have been extensively used in the previous studies.19, 21, 36 We were able to investigate temporality of associations between exposure, outcomes, and confounders—a necessary precondition for causal inference. Measuring self-reported disordered eating behaviours in the general population also has the advantage of minimising selection biases associated with the use of clinical cases (because only a few individuals with eating disorders reach clinical services),40 as well as capturing more common phenotypical presentations of otherwise rare clinical disorders. Finally, our dataset covered a substantial amount of the adolescent period, during which most disordered eating behaviours tend to emerge.8

(because only a few individuals with eating disorders reach clinical services),40 as well as capturing more common phenotypical presentations of otherwise rare clinical disorders. Finally, our dataset covered a substantial amount of the adolescent period, during which most disordered eating behaviours tend to emerge.8 Despite these strengths, our study also had several limitations. Although disordered-eating behaviours are more common than diagnoses, only a few adolescents reported these outcomes in our dataset. This low number of reports could have resulted in low statistical power and increased possibility of types I and II errors, especially in complete case analyses and when investigating individual behaviours and particularly when running multiple statistical tests. We saw similar effect sizes, but more precise CIs when we pooled all disordered eating behaviours together (increasing our statistical power). This finding suggests that the weak associations observed for individual outcomes could indeed have resulted from low power.

articularly when running multiple statistical tests. We saw similar effect sizes, but more precise CIs when we pooled all disordered eating behaviours together (increasing our statistical power). This finding suggests that the weak associations observed for individual outcomes could indeed have resulted from low power. Our disordered eating definitions were broad. Some adolescents included in our outcome definition might therefore have had very low threshold behaviours (ie, occurring less frequently than that required by diagnostic manuals for a diagnosis of an eating disorder), which could have artificially diluted associations. To overcome this limitation and to investigate the association between psychotic experiences and more severe eating-disorder psychopathology, we created a variable indicating the number of co-occurring disordered-eating behaviours. We noted that children with psychotic experiences also reported more severe disordered eating outcomes. Our findings therefore appear to be consistent with the possibility that, like other disorders, eating disorders exist on a continuum of severity from disordered eating to diagnosed-eating disorders.

ting behaviours. We noted that children with psychotic experiences also reported more severe disordered eating outcomes. Our findings therefore appear to be consistent with the possibility that, like other disorders, eating disorders exist on a continuum of severity from disordered eating to diagnosed-eating disorders. It is possible that losses to follow up could have resulted in some degree of selection bias. To account for this bias, we imputed missing covariate and outcome data, and with the exception of purging behaviours (for which we observed stronger associations in complete case analyses), the results were similar. Substance use could have been a potential confounder of the association between psychotic experiences and disordered eating. However, substance use before age 13 years was rare in our sample, so we did not have enough power to control for it; future studies should aim to include these behaviours. Our measure of BMI did not account for pubertal status (as acknowledged by the researchers who developed this methodology).32 We did not adjust for pubertal timing at baseline because we hypothesised in our direct acyclical graph that it would exert its effect on the outcome via BMI and depressive symptoms at baseline (appendix). However, this assumption relied on our hypothesised risk mechanisms and might have missed other biological and psychological mechanisms, which should be investigated in future studies.

sed in our direct acyclical graph that it would exert its effect on the outcome via BMI and depressive symptoms at baseline (appendix). However, this assumption relied on our hypothesised risk mechanisms and might have missed other biological and psychological mechanisms, which should be investigated in future studies. Finally, we were not able to measure excessive concerns with healthy eating behaviours, which might be a marker of orthorexia nervosa, because these outcomes are not measured by the Youth Risk Behaviour Surveillance System questions included in ALSPAC. Given increasing public health concerns around these behaviours, future studies should also aim to collect this information. We noted that children reporting psychotic experiences at age 13 years were more likely to have any disordered eating behaviours in adolescence with more severe symptoms, defined as a greater number of behaviours. These results align well with those from previous investigations showing that adolescents and adults with psychotic experiences are more likely to also report co-occurring disordered eating behaviours3 and develop diagnosed eating disorders,11 and those with eating disorders are more likely to have12 or subsequently report psychotic symptoms.11 However, these studies—given their cross-sectional4, 12 or retrospective nature11—could not clearly establish the temporality of these associations4, 12 or might have been affected by recall bias.11 The prospective nature of our investigation, on the other hand, allowed us to show for the first time that psychotic experiences are likely to precede the onset of disordered-eating behaviours, especially since disordered-eating behaviours tend to be rare before age 13 years and only typically emerge in mid-to-late adolescence.8

ctive nature of our investigation, on the other hand, allowed us to show for the first time that psychotic experiences are likely to precede the onset of disordered-eating behaviours, especially since disordered-eating behaviours tend to be rare before age 13 years and only typically emerge in mid-to-late adolescence.8 Similar to results from a previous study of adults that showed that those with psychotic symptoms had greater odds of subsequently developing bulimia nervosa,11 we noted that adolescents who reported psychotic experiences at age 13 years had greater odds in both crude models and adjusted models 1 to 3 of reporting purging and binge eating behaviours at age 18 years—core symptoms of bulimia nervosa. Binge eating is also a core symptom of binge-eating disorder for which a lifetime comorbidity with psychotic symptoms also occurs.11 Fasting is a restrictive behaviour which, although typical of anorexia nervosa, might also represent a non-purging compensatory behaviour in bulimia nervosa, and has been shown to be a strong predictor of onset of bingeing and purging behaviours in adolescence.41 Crossover between diagnoses of eating disorders is common and estimated to occur in up to 50% of cases between anorexia nervosa and bulimia nervosa.42 Hence, it is perhaps unsurprising that we observed an effect across the whole eating disorders behavioural spectrum as opposed to finding specific associations.

cence.41 Crossover between diagnoses of eating disorders is common and estimated to occur in up to 50% of cases between anorexia nervosa and bulimia nervosa.42 Hence, it is perhaps unsurprising that we observed an effect across the whole eating disorders behavioural spectrum as opposed to finding specific associations. Eating disorders, particularly bulimia nervosa and binge eating disorder,8 are highly comorbid with mood disorders. In our analyses, adjustment for depressive symptoms attenuated the magnitude of the association between psychotic experiences and disordered eating and in some cases—eg, binge eating—completely accounted for the association. This finding suggests that transdiagnostic psychopathologies in early adolescence could increase future disordered eating behaviours. A study in this sample7 has shown that depressive and anxiety symptoms, and psychotic experiences occur on a continuum of severity with psychotic experiences denoting more acute psychopathological presentations rather than being distinct diagnostic entities. Psychotic experiences could therefore represent one set of early markers of more severe mental illness with transdiagnostic effects that extend, among other conditions,3, 5, 6 to eating disorders. The presence of shared risk factors between depression, psychosis, and eating disorders—such as bullying43, 44 or trauma45, 46—seems to corroborate this hypothesis.

resent one set of early markers of more severe mental illness with transdiagnostic effects that extend, among other conditions,3, 5, 6 to eating disorders. The presence of shared risk factors between depression, psychosis, and eating disorders—such as bullying43, 44 or trauma45, 46—seems to corroborate this hypothesis. Adjustment for depressive symptoms attenuated, but did not completely explain the association between psychotic experiences and purging, fasting, and any or more severe disordered eating. It is therefore possible that the association between psychotic experiences and eating disorders could also be underpinned by specific shared risk factors, such as cognitive impairments. For instance, individuals with either eating disorders or schizophrenia show similar deficits in visuospatial and global or local processing, with greater bias towards detail processing.47, 48 Most research into eating disorders has been done in clinical samples, which makes it difficult to disentangle whether cognitive deficits precede or are a consequence of eating disorders. Our finding that psychotic experiences, previously associated with several cognitive abnormalities,19, 22 are longitudinally associated with disordered eating suggests that certain cognitive impairments could precede the development of eating disorders.

gnitive deficits precede or are a consequence of eating disorders. Our finding that psychotic experiences, previously associated with several cognitive abnormalities,19, 22 are longitudinally associated with disordered eating suggests that certain cognitive impairments could precede the development of eating disorders. Clinical accounts of eating disorders also lend plausibility to the hypothesis of commonalities across psychotic and eating disorders. For instance, it has been suggested that for a subgroup of people with concerns about their weight and shape, such symptoms could resemble delusions typically observed in psychosis.49 Individuals with eating disorders also recount experiences of hearing an inner voice,50 which can become progressively more hostile, controlling, and judgmental of their eating behaviours, weight, and shape. Unlike psychosis, however, these occurrences are not commonly understood as external voices.51 More longitudinal research exploring these putative shared risk mechanisms is therefore warranted.47

e,50 which can become progressively more hostile, controlling, and judgmental of their eating behaviours, weight, and shape. Unlike psychosis, however, these occurrences are not commonly understood as external voices.51 More longitudinal research exploring these putative shared risk mechanisms is therefore warranted.47 Findings from genetic studies also support the hypothesis that psychotic and eating disorders could have shared causal pathways. Several investigations using the results of large genome-wide association studies have uncovered a positive genetic correlation between schizophrenia and anorexia nervosa, and a negative correlation between both disorders and BMI.52, 53 These results suggest that schizophrenia and anorexia nervosa might be characterised by shared metabolic and developmental risk pathways. For instance, in the ALSPAC cohort, polygenic risk scores for schizophrenia have been shown to be associated with deficits in social communication,54 also proposed as a risk factor for eating disorders, as well as anxiety.8, 55, 56

exia nervosa might be characterised by shared metabolic and developmental risk pathways. For instance, in the ALSPAC cohort, polygenic risk scores for schizophrenia have been shown to be associated with deficits in social communication,54 also proposed as a risk factor for eating disorders, as well as anxiety.8, 55, 56 There were no associations between psychotic experiences and BMI in our study. There is evidence that underweight BMI is a risk factor for schizophrenia.57 However, we observed that psychotic experiences were associated with greater binge-eating behaviours, shown to predict greater BMI in this population,58 and this finding requires further investigation. Greater binge eating behaviours in children with psychotic experiences, however, could explain the increased rates of metabolic disorder observed in drug-naive individuals with psychosis, since binge eating has been proposed as a risk factor for metabolic abnormalities independently of obesity.59 In our sample, psychotic experiences at 18 years have also recently been linked to increased rates of insulin resistance,60 also occurring in individuals who binge eat61 and a feature of diabetes, often comorbid with eating disorders.62 More research exploring these shared metabolic risk pathways is therefore warranted.

n our sample, psychotic experiences at 18 years have also recently been linked to increased rates of insulin resistance,60 also occurring in individuals who binge eat61 and a feature of diabetes, often comorbid with eating disorders.62 More research exploring these shared metabolic risk pathways is therefore warranted. In conclusion, we showed that psychotic experiences in early adolescence were longitudinally associated with disordered eating behaviours at age 18 years, potentially with more severe presentations. Our findings provide further evidence that psychotic experiences could represent non-specific markers of adverse psychopathology in adolescence1, 11 and extend their relevance to disordered eating behaviours, previously largely overlooked by the scientific literature. Our study, in the context of broader clinical and genetic studies, supports the hypothesis of shared causal pathways between psychotic illness and eating disorders, which requires further investigation. Finally, our results suggest that increased disordered behaviours, especially cycles of binge-eating and fasting, could partly account for increased rates of metabolic abnormalities in individuals with psychotic illness,63 although more research investigating these mechanisms is needed. For more on ALSPAC see http://www.bris.ac.uk/alspac/researchers/data-access/data-dictionary/ Supplementary Material Supplementary appendix

In conclusion, we showed that psychotic experiences in early adolescence were longitudinally associated with disordered eating behaviours at age 18 years, potentially with more severe presentations. Our findings provide further evidence that psychotic experiences could represent non-specific markers of adverse psychopathology in adolescence1, 11 and extend their relevance to disordered eating behaviours, previously largely overlooked by the scientific literature. Our study, in the context of broader clinical and genetic studies, supports the hypothesis of shared causal pathways between psychotic illness and eating disorders, which requires further investigation. Finally, our results suggest that increased disordered behaviours, especially cycles of binge-eating and fasting, could partly account for increased rates of metabolic abnormalities in individuals with psychotic illness,63 although more research investigating these mechanisms is needed. For more on ALSPAC see http://www.bris.ac.uk/alspac/researchers/data-access/data-dictionary/ Supplementary Material Supplementary appendix Acknowledgments JBK and FS are supported by a Sir Henry Dale Fellowship to JBK, jointly funded by the Wellcome Trust and the Royal Society (number 101272/Z/13/Z). FS was also supported by a Sir Henry Dale Wellcome Fellowship funded by the Wellcome Trust (number 209196/Z/17/Z). GL is supported by the UCLH NIHR Biomedical Research Centre. The UK Medical Research Council, Wellcome (number 102215/2/13/2), and the University of Bristol provide core support for the Avon Longitudinal Study of Parents and Children (ALSPAC). We are grateful to all the families who took part in this study, the midwives for their help in recruiting them, and the whole ALSPAC team, which includes interviewers, computer and laboratory technicians, clerical workers, research scientists, volunteers, managers, receptionists and nurses. We also thank Gemma Lewis for her invaluable feedback on the manuscript.

part in this study, the midwives for their help in recruiting them, and the whole ALSPAC team, which includes interviewers, computer and laboratory technicians, clerical workers, research scientists, volunteers, managers, receptionists and nurses. We also thank Gemma Lewis for her invaluable feedback on the manuscript. Contributors FS conceptualised the research question and developed the analysis plan jointly with all authors. FS analysed the data and drafted the manuscript. DM contributed to a first stage of data analysis and manuscript drafting. DM, GL, and JBK read and contributed to revising the manuscript. JBK provided senior supervision for the manuscript. Declaration of interests We declare no competing interests.

Introduction Wilms tumour is a childhood kidney tumour that affects one in 10 000 children. Its histology is similar to that of the developing kidney and is typically triphasic, with blastemal, stromal, and epithelial components.1 Biphasic tumours, with two of the three components, and monomorphic tumours consisting of only one component also occur but are rarer.1

y tumour that affects one in 10 000 children. Its histology is similar to that of the developing kidney and is typically triphasic, with blastemal, stromal, and epithelial components.1 Biphasic tumours, with two of the three components, and monomorphic tumours consisting of only one component also occur but are rarer.1 Wilms tumour is primarily a non-familial condition, with only about 2% of affected individuals having a relative with the tumour.2 Given its rarity, inherited causes—rather than chance—are assumed to underlie familial clusters, and several have been reported, including constitutional mutations in the genes WT1, CTR9, and REST.2, 3, 4 Wilms tumour is also known to be associated with many genetic conditions, including the WAGR, Denys-Drash, Beckwith-Wiedemann, Simpson-Golabi-Behmel, Perlman, mosaic variegated aneuploidy, hereditary hyperparathyroidism-jaw tumour, Li-Fraumeni, DICER1, and Bohring-Opitz syndromes, Fanconi anaemia, and PIK3CA-related overgrowth spectrum.2, 3, 4, 5, 6, 7, 8, 9, 10, 11 These conditions are diverse in their clinical and histological associations, inheritance patterns, and mutational mechanisms of pathogenicity. The underlying predisposition genes have many different functions and are involved in diverse biological processes. The identification of these genes and investigations into their role in Wilms tumour predisposition have led to fundamental insights into developmental, cellular, and oncological mechanisms and have important clinical implications for individuals with Wilms tumour and their families.2, 3, 4, 5, 6, 7, 8, 9, 10, 11 Furthermore, many causes of familial and syndromic Wilms tumour also contribute to non-familial, non-syndromic Wilms tumour.2, 4, 5

sights into developmental, cellular, and oncological mechanisms and have important clinical implications for individuals with Wilms tumour and their families.2, 3, 4, 5, 6, 7, 8, 9, 10, 11 Furthermore, many causes of familial and syndromic Wilms tumour also contribute to non-familial, non-syndromic Wilms tumour.2, 4, 5 Research in context Evidence before this study Wilms tumour is a rare childhood kidney tumour. We searched PubMed for papers in English with the terms “Wilms” AND “genetic” OR “mutation” OR “familial” OR “syndrome”, yielding 2801 papers that we reviewed to identify those relevant to genetic predisposition to Wilms tumour. This review identified 17 genes previously shown to predispose to Wilms tumour and showed that further Wilms tumour predisposition genes must exist, because many syndromic cases and familial clusters have not been explained. Added value of this study

Wilms tumour is a rare childhood kidney tumour. We searched PubMed for papers in English with the terms “Wilms” AND “genetic” OR “mutation” OR “familial” OR “syndrome”, yielding 2801 papers that we reviewed to identify those relevant to genetic predisposition to Wilms tumour. This review identified 17 genes previously shown to predispose to Wilms tumour and showed that further Wilms tumour predisposition genes must exist, because many syndromic cases and familial clusters have not been explained. Added value of this study To our knowledge, this study is the largest exome sequencing study to date of individuals with Wilms tumour, involving 890 individuals, including 91 individuals from 49 familial Wilms tumour pedigrees. We identified four new Wilms tumour predisposition genes, TRIM28, FBXW7, KDM3B, and NYNRIN. We showed that FBXW7 and KDM3B are pleiotropic cancer predisposition genes, and that KDM3B and NYNRIN might also cause non-malignant phenotypes, particularly intellectual disability. Our study identified TRIM28 as a major Wilms tumour predisposition gene, making a similar contribution to familial and unselected Wilms tumour as those of constitutional WT1 and REST mutations. We also found an association between TRIM28 mutations and epithelial histology and a strong parent-of-origin-effect, because all inherited TRIM28 mutations were maternally transmitted. Functional enrichment analyses revealed remarkable diversity in the biological pathways affected by Wilms tumour predisposition genes. We also found limited overlap between the 21 constitutionally mutated Wilms tumour predisposition genes and 20 genes somatically mutated in Wilms tumour.

ns were maternally transmitted. Functional enrichment analyses revealed remarkable diversity in the biological pathways affected by Wilms tumour predisposition genes. We also found limited overlap between the 21 constitutionally mutated Wilms tumour predisposition genes and 20 genes somatically mutated in Wilms tumour. Implications of all the available evidence This study provides new insights into the causes of Wilms tumour and describes the overall landscape of Wilms tumour predisposition. Wilms tumour shows remarkable genetic heterogeneity and aetiological complexity, which have substantial clinical impact. Genetic testing should be made available to individuals with Wilms tumour, but will need to encompass both broad genetic testing, for example by exome sequencing, and testing for 11p15 epigenetic abnormalities. Moreover, our findings suggest that more Wilms tumour predisposition genes are likely to exist, which will have relevance for future research and clinical testing. Research over the past 25 years has led to tremendous advances in our knowledge of Wilms tumour predisposition. However, available evidence suggests that our knowledge is still incomplete and that further predisposition factors remain to be discovered. In particular, the cause of many familial clusters is still unknown.4 In this study, we aimed to use exome sequencing to identify new Wilms tumour predisposition genes and to characterise and contextualise the genetic landscape of such predisposition.

ete and that further predisposition factors remain to be discovered. In particular, the cause of many familial clusters is still unknown.4 In this study, we aimed to use exome sequencing to identify new Wilms tumour predisposition genes and to characterise and contextualise the genetic landscape of such predisposition. Methods Study design and participants In this exome sequencing study, we included participants recruited to the Factors Associated with Childhood Tumours (FACT) Study. All children with a childhood solid tumour from the UK were eligible for participation in the FACT study; children with familial childhood cancer anywhere in the world were also eligible for participation in the FACT study. We analysed lymphocyte DNA from 1215 individuals with 28 different childhood tumours, of whom 1206 had one childhood tumour and nine individuals had two different childhood tumours (appendix). This cohort included 890 individuals with Wilms tumour: 799 had non-familial disease and 91 were from 49 familial Wilms tumour pedigrees in which two or more individuals had Wilms tumour due to an unknown genetic or epigenetic cause (figure 1; appendix). Most participants with Wilms tumour were from the UK and, therefore, were likely to have been treated with chemotherapy before surgery. We used constitutional (germline) exome data from 7632 individuals with 28 different adult cancers available from The Cancer Genome Atlas (TCGA) on May 13, 2014 (figure 1; see appendix for the types and number of adult cancers interrogated). As reference data, we used the Exome Aggregation Consortium (ExAC) data, version 3, accessed on Nov 13, 2015 (excluding the TCGA samples),12 and the ICR1000 UK exome series.13 We generated and analysed the ICR1000 UK exome series and childhood cancer sample data from the FACT participants by use of consistent sequencing and analytical processes.Figure 1 Cancer cohorts investigated

AC) data, version 3, accessed on Nov 13, 2015 (excluding the TCGA samples),12 and the ICR1000 UK exome series.13 We generated and analysed the ICR1000 UK exome series and childhood cancer sample data from the FACT participants by use of consistent sequencing and analytical processes.Figure 1 Cancer cohorts investigated Wilms tumour families are pedigrees in which two or more individuals had Wilms tumour. FACT=Factors Associated with Childhood Tumours. TCGA=The Cancer Genome Atlas. *Nine individuals had two different childhood tumours. The FACT study was approved by the London Multicentre Ethics Committee (05/MRE02/17), and its collaborators are listed in the appendix. Written informed consent was obtained from all participants, their parents or guardians, or both, as appropriate (age cutoff for consent was 18 years).

Wilms tumour families are pedigrees in which two or more individuals had Wilms tumour. FACT=Factors Associated with Childhood Tumours. TCGA=The Cancer Genome Atlas. *Nine individuals had two different childhood tumours. The FACT study was approved by the London Multicentre Ethics Committee (05/MRE02/17), and its collaborators are listed in the appendix. Written informed consent was obtained from all participants, their parents or guardians, or both, as appropriate (age cutoff for consent was 18 years). Procedures We did exome sequencing in samples from all childhood cancer probands and 119 individuals from 49 familial Wilms tumour pedigrees who did not have Wilms tumour, by using 50 ng genomic DNA and the Nextera DNA sample preparation kit (Illumina, San Diego, CA, USA) or 1·5 μg genomic DNA and the TruSeq exome enrichment kit (Illumina). The captured libraries were amplified by PCR with the supplied paired-end PCR primers. Sequencing was done with HiSeq 2000 (Illumina) or HiSeq 2500 (Illumina). We used the OpEx v1.0 pipeline to do variant calling in childhood cancer, adult cancer (TGCA), and ICR1000 exome data.14 We also reannotated the variants in the ExAC data with the CAVA tool in OpEx, to ensure variant calling consistency across the different cohorts.14 We used the protein-truncating variant prioritisation method to prioritise potential disease-associated genes for follow-up; this is a proven strategy for identifying tumour suppressor genes in outbred populations, which we have used to identify several other cancer predisposition genes.3, 4, 10 We validated variants in TRIM28, FBXW7, NYNRIN, and KDM3B genes by use of Sanger sequencing in the probands and any available relatives, designing primers with BatchPrimer3. We used the QIAGEN Multiplex PCR kit (QIAGEN, Hilden, Germany) to prepare PCRs, and the resulting amplicons were bidirectionally sequenced with BigDye Terminator cycle sequencing kits (Thermo Fisher Scientific, Waltham, MA, USA) and an ABI 3730 sequencer (Life Technologies, Carlsbad, CA, USA). We analysed sequencing traces with Mutation Surveyor software and by visual inspection with Chromas, version 2.13. We validated the CDC73 mutation with the TruSight Cancer panel (Illumina). We did in-silico analyses of variant pathogenicity with Alamut Visual, version 2.9.0.

ncer (Life Technologies, Carlsbad, CA, USA). We analysed sequencing traces with Mutation Surveyor software and by visual inspection with Chromas, version 2.13. We validated the CDC73 mutation with the TruSight Cancer panel (Illumina). We did in-silico analyses of variant pathogenicity with Alamut Visual, version 2.9.0. Statistical analysis We used the methods described by Akawi and colleagues15 to obtain the probability of a family in our study having two protein-truncating variants in a given gene. The method uses the frequency of rare protein-truncating variants (allele frequency <0·001) in ExAC and the number of observed protein-truncating variants in a given gene to estimate the probability of an individual having two of these variants in that gene. The baseline prevalence of having two protein-truncating variants per gene is calculated as the proportion of rare protein-truncating variants squared. We observed two individuals with two protein-truncating variants and nine individuals with a single protein-truncating variant in NYNRIN among 844 individuals with Wilms tumour from the 890 included in this study. We used the R function to calculate the probability of observing two individuals with two NYNRIN protein-truncating variants: analyse_inherited_enrichment from the R package recessiveStats with hgnc=”NYNRIN”, chrom=“14”, counts$biallelic_lof=2, counts$lof_func=9, and cohort_n=844.

r from the 890 included in this study. We used the R function to calculate the probability of observing two individuals with two NYNRIN protein-truncating variants: analyse_inherited_enrichment from the R package recessiveStats with hgnc=”NYNRIN”, chrom=“14”, counts$biallelic_lof=2, counts$lof_func=9, and cohort_n=844. We used a binomial test—dbinom function in R—to calculate the probability of all ten TRIM28 mutations with known inheritance being maternally inherited, assuming the baseline probability of inheriting the variant from either parent was 0·5. We did a functional enrichment analysis with use of g:Profiler (version r1665_e85_eg32).16 We used the 21 predisposition genes described for Wilms tumour as our query set. We looked for enrichment among Gene Ontology molecular function terms and pathway gene sets from the Kyoto Encyclopedia of Genes and Genomes, requiring the size of the functional category to have a minimum of five genes and using the Benjamini-Hochberg correction for multiple testing p value as the significance threshold. The false discovery rate q values presented in this study are the Benjamini-Hochberg critical values.

o Encyclopedia of Genes and Genomes, requiring the size of the functional category to have a minimum of five genes and using the Benjamini-Hochberg correction for multiple testing p value as the significance threshold. The false discovery rate q values presented in this study are the Benjamini-Hochberg critical values. For the somatic cancer driver comparisons, we used 20 genes reported to be somatically mutated in Wilms tumour. These included 17 established genes reported in more than one publication and three newly reported genes (ACTB, BCOR, NONO) with at least three somatic mutations in the TARGET discovery series.17 We used the COSMIC cancer gene census to establish which of the 21 Wilms tumour predisposition genes, and which of the 20 somatically mutated Wilms tumour driver genes, were also somatically mutated in other cancer types. Role of the funding source The funder of the study had no role in study design, data collection, data analysis, data interpretation, or writing of the report. SM, SY, EH, and NR had full access to all the data. The corresponding author had full access to all the data in the study and had final responsibility for the decision to submit for publication.

nder of the study had no role in study design, data collection, data analysis, data interpretation, or writing of the report. SM, SY, EH, and NR had full access to all the data. The corresponding author had full access to all the data in the study and had final responsibility for the decision to submit for publication. Results We used the protein-truncating variant prioritisation method with dominant and recessive inheritance models to identify genes with different protein-truncating variants in 890 individuals with Wilms tumour (figure 1). The three genes with the strongest signal in the prioritisation analyses were TRIM28, FBXW7, and NYNRIN. We did Sanger sequencing to validate protein-truncating variants and rare non-synonymous variants of these genes in probands and any available samples from relatives to further evaluate their status as bona fide Wilms tumour predisposition genes.

est signal in the prioritisation analyses were TRIM28, FBXW7, and NYNRIN. We did Sanger sequencing to validate protein-truncating variants and rare non-synonymous variants of these genes in probands and any available samples from relatives to further evaluate their status as bona fide Wilms tumour predisposition genes. We identified pathogenic truncating mutations of TRIM28 in 17 individuals with Wilms tumour from 13 families (figure 2, table, appendix). At least three of these truncating mutations had arisen de novo. Protein-truncating variants in TRIM28 are extremely rare in the general population, because the gene is highly intolerant to truncating variation, with a pLI score of 1·0 (pLI >0·9 indicates extreme intolerance to protein-truncating variants).12 We found no other cancers in individuals carrying TRIM28 mutations. We also did not find any TRIM28 protein-truncating variants in 334 individuals with 27 other childhood cancers or in 7632 individuals with adult cancers, suggesting that TRIM28 pathogenic mutations primarily predispose to Wilms tumour. Of note, the TRIM28 mutations in two families in our study (ID_0498 and ID_0506) were independently, and coincidentally, reported while we were preparing this manuscript.18 We identified a de-novo stop-loss mutation in another family (ID_7574), which we assumed to be pathogenic. Finally, in family ID_0477, which included six cases of Wilms tumour, we identified TRIM28 929G→A, leading to the protein change Gly310Asp (figure 2, table, appendix). We believe this mutation to be pathogenic because it is absent from public and in-house datasets, it segregates with Wilms tumour in the family, it is predicted to be deleterious by in silico tools, and it is at a crucial residue within the coiled-coil domain of TRIM28 that is reported to interact with AMER1, which is encoded by a gene somatically mutated in Wilms tumour.19Figure 2 Schematic representations of TRIM28, FBXW7, NYNRIN, and KDM3B

our in the family, it is predicted to be deleterious by in silico tools, and it is at a crucial residue within the coiled-coil domain of TRIM28 that is reported to interact with AMER1, which is encoded by a gene somatically mutated in Wilms tumour.19Figure 2 Schematic representations of TRIM28, FBXW7, NYNRIN, and KDM3B Schematic representations of encoded proteins are shown, with functional domains in grey. The position of cancer-predisposing mutations is shown above the protein. Red symbols denote de novo mutations. Table Molecular and clinical features of individuals with mutations in TRIM28, FBXW7, NYNRIN, KDM3B, or CDC73

our in the family, it is predicted to be deleterious by in silico tools, and it is at a crucial residue within the coiled-coil domain of TRIM28 that is reported to interact with AMER1, which is encoded by a gene somatically mutated in Wilms tumour.19Figure 2 Schematic representations of TRIM28, FBXW7, NYNRIN, and KDM3B Schematic representations of encoded proteins are shown, with functional domains in grey. The position of cancer-predisposing mutations is shown above the protein. Red symbols denote de novo mutations. Table Molecular and clinical features of individuals with mutations in TRIM28, FBXW7, NYNRIN, KDM3B, or CDC73 Sex Mutations (protein change) Inheritance Tumour type, age at diagnosis (months) Unilateral or bilateral Histology Other Status, age (years) TRIM28 ID_0477_01 F 929G→A (Gly310Asp) Maternal WT, 24 Unilateral Epithelial predominant .. .. ID_0477_02 M 929G→A (Gly310Asp) Maternal WT, 84 Unilateral Epithelial .. .. ID_0477_03 F 929G→A (Gly310Asp) Maternal WT, 93 Unilateral NA .. .. ID_0498_01 M 1746_1747delinsC Maternal WT, 8 Unilateral Epithelial .. Alive, 30 ID_0498_02 F 1746_1747delinsC Maternal WT, 5 Unilateral Epithelial .. Alive, 29 ID_0498_03 F 1746_1747delinsC NA WT, 6 Unilateral Epithelial .. .. ID_0487_01 M 429dupC Maternal WT, 15 Unilateral Epithelial predominant .. Alive, 19 ID_0487_02 M 429dupC NA WT, 18 Unilateral NA .. .. ID_0506_01 M 525_526delGA Maternal WT, 39 Unilateral Epithelial .. Alive, 23 ID_0506_02 F 525_526delGA Maternal WT, 8 Bilateral Epithelial .. Alive, 20 ID_7487_01 F 239_245del7 Maternal WT, 118 Unilateral Epithelial predominant, diffuse anaplasia .. Died, 12 ID_1982 M 1957delC De novo WT, 11 Bilateral Epithelial predominant .. Alive, 15 ID_6530 M 209_210delAG De novo WT, 15 Unilateral Epithelial and blastemal Autism, speech delay Alive, 6 ID_1969 M 840–2A→G De novo WT, 118 Unilateral Epithelial and blastemal .. Alive, 19 ID_7574 M 2508A→G (X836TrpextX?)* De novo WT, 13 Unilateral Epithelial predominant Autism, intellectual disability .. ID_0902 F 1250C→A (Ser417X) Maternal WT, 12 Unilateral Epithelial predominant .. .. ID_0692 F 1459C→T (Arg487X) NA WT, 13 Bilateral NA .. Alive, 36 ID_6671 F 688C→T (Arg230X) NA WT, 10 Bilateral Epithelial predominant Chronic kidney disease Alive, 6 ID_0796 F 1085T→A (Leu362X) NA WT, 61 Unilateral NA .. Alive, 33 ID_0866 F 1300_1301dupAA NA WT, 90 Unilateral Epithelial predominant .. Alive, 29 ID_0936 M 1150G→T (Glu384X) NA WT, 8 Unilateral NA .. .. FBXW7 ID_0811 M 710G→A (Trp237X) De novo WT, 76 Unilateral NA Osteosarcoma at 39 years Died, 39 ID_2084_01 M 1972C→T (Arg658X) NA WT, 42 Unilateral Focal anaplasia Relapse at 66 months .. ID_0592 F 1017_1021del5 Paternal WT, 28 Unilateral NA hypotonia Alive, 18 ID_1227 F 670C→T (Arg224X) NA WT, 73 .. NA ..

ral NA .. .. FBXW7 ID_0811 M 710G→A (Trp237X) De novo WT, 76 Unilateral NA Osteosarcoma at 39 years Died, 39 ID_2084_01 M 1972C→T (Arg658X) NA WT, 42 Unilateral Focal anaplasia Relapse at 66 months .. ID_0592 F 1017_1021del5 Paternal WT, 28 Unilateral NA hypotonia Alive, 18 ID_1227 F 670C→T (Arg224X) NA WT, 73 .. NA .. Died, 7 ID_7520 M 1753A→T (Ser585Cys) De novo Rhabdoid, 40 .. Extra-renal rhabdoid with INI1 loss Two febrile convulsions Alive, 5 NYNRIN ID_0493_01 M 1955_1956delCA Paternal WT, 24 Unilateral Blastemal predominant Inguinal hernia .. ID_0493_01 .. 3761G→A (Trp1254X) Maternal .. .. .. .. .. ID_0493_02 M 1955_1956delCA Paternal WT, 24 Unilateral Triphasic .. .. ID_0493_02 .. 3761G→A (Trp1254X) Maternal .. .. .. .. .. ID_6049 M 311G→A (Trp104X) Maternal WT, 34 Unilateral Triphasic Epilepsy, hypothyroidism, intellectual disability Alive, 11 ID_6049 .. 1295_1296del2ins31 Paternal .. .. .. .. .. KDM3B ID_7225 F 3422A→G (Asn1141Ser) De novo WT, 49 Bilateral NA Hyperpigmentation .. ID_2086 M 916_917delAG De novo Hepatoblastoma, 131 .. NA Autism, abnormal pigmentation, intellectual disability .. CDC73 ID_6491_01 F 878dupA Paternal WT, 192 Unilateral Epithelial predominant Convergent strabismus Alive, 21 ID_6491_02 M 878dupA NA WT, 96 Unilateral NA .. Alive, 48 Pedigrees and chromatograms are shown in the appendix. F=female. WT=Wilms tumour. M=male. NA=not available. * The stop codon (X) at position 836 is changed to Trp, extending the protein by an unknown number of amino acids (?).

Died, 7 ID_7520 M 1753A→T (Ser585Cys) De novo Rhabdoid, 40 .. Extra-renal rhabdoid with INI1 loss Two febrile convulsions Alive, 5 NYNRIN ID_0493_01 M 1955_1956delCA Paternal WT, 24 Unilateral Blastemal predominant Inguinal hernia .. ID_0493_01 .. 3761G→A (Trp1254X) Maternal .. .. .. .. .. ID_0493_02 M 1955_1956delCA Paternal WT, 24 Unilateral Triphasic .. .. ID_0493_02 .. 3761G→A (Trp1254X) Maternal .. .. .. .. .. ID_6049 M 311G→A (Trp104X) Maternal WT, 34 Unilateral Triphasic Epilepsy, hypothyroidism, intellectual disability Alive, 11 ID_6049 .. 1295_1296del2ins31 Paternal .. .. .. .. .. KDM3B ID_7225 F 3422A→G (Asn1141Ser) De novo WT, 49 Bilateral NA Hyperpigmentation .. ID_2086 M 916_917delAG De novo Hepatoblastoma, 131 .. NA Autism, abnormal pigmentation, intellectual disability .. CDC73 ID_6491_01 F 878dupA Paternal WT, 192 Unilateral Epithelial predominant Convergent strabismus Alive, 21 ID_6491_02 M 878dupA NA WT, 96 Unilateral NA .. Alive, 48 Pedigrees and chromatograms are shown in the appendix. F=female. WT=Wilms tumour. M=male. NA=not available. * The stop codon (X) at position 836 is changed to Trp, extending the protein by an unknown number of amino acids (?). We established that ten of the TRIM28 mutations had been inherited, and that in all cases, the mutation had been transmitted from the mother, a significant association (p=0·00098). Pathology information was available for 16 tumours, of which 14 were epithelial or epithelial-predominant (table).

* The stop codon (X) at position 836 is changed to Trp, extending the protein by an unknown number of amino acids (?). We established that ten of the TRIM28 mutations had been inherited, and that in all cases, the mutation had been transmitted from the mother, a significant association (p=0·00098). Pathology information was available for 16 tumours, of which 14 were epithelial or epithelial-predominant (table). We identified truncating FBXW7 mutations in four individuals with Wilms tumour, of which one was de novo (in ID_0811), one had been inherited from an unaffected father (in ID_0592), and two were of unknown provenance (in ID_2084 and ID_1227; figure 2, table, appendix). FBXW7 is highly intolerant to protein-truncating variants (pLI=1·00) and these data suggest that FBXW7 is a Wilms tumour predisposition gene. Two of the four individuals with truncating FBXW7 mutations have died (table). Additionally, ID_2084 was treated for Wilms tumour at 3·5 years of age, but relapsed when he was 5·5 years old. We did not find truncating FBXW7 mutations in individuals with other childhood or adult cancers. However, we identified a de novo non-synonymous mutation, 1753A→T (protein change Ser585Cys), in a child with an extra-renal rhabdoid tumour (ID_7520), which we assumed to be pathogenic.

relapsed when he was 5·5 years old. We did not find truncating FBXW7 mutations in individuals with other childhood or adult cancers. However, we identified a de novo non-synonymous mutation, 1753A→T (protein change Ser585Cys), in a child with an extra-renal rhabdoid tumour (ID_7520), which we assumed to be pathogenic. We identified biallelic truncating mutations in NYNRIN in three children from two families (ID_0493 and ID_6049; figure 2, table, appendix). Each parent was heterozygous for one of the mutations. These mutations were absent from ExAC and the ICR1000 series. We found no individuals with two NYNRIN truncating mutations in the ICR1000 series and no homozygous protein-truncating variants in ExAC (individual-level data is not available for ExAC, therefore it is not possible to know if anyone had two different protein-truncating variants). Additionally, the probability of finding two different families with the same phenotype and two truncating NYNRIN mutations by chance is 4·0 × 10−9. One of the affected children had an inguinal hernia and another had epilepsy, hypothyroidism, and intellectual disability (table). It is unclear whether any of these additional clinical features are related to the biallelic NYNRIN mutations. We did not identify biallelic NYNRIN protein-truncating variants in individuals with other childhood or adult cancers.

inal hernia and another had epilepsy, hypothyroidism, and intellectual disability (table). It is unclear whether any of these additional clinical features are related to the biallelic NYNRIN mutations. We did not identify biallelic NYNRIN protein-truncating variants in individuals with other childhood or adult cancers. In addition to the agnostic protein-truncating variant prioritisation analyses, we reviewed the exome data in genes proposed as possible childhood cancer predisposition genes, identified through a systematic review of 19 171 genes for links to Mendelian disease. This review led us to the identification of two de-novo KDM3B mutations, a non-synonymous mutation in a child with Wilms tumour and a hyperpigmented lesion on her buttock (ID_7225) and a truncating mutation in a child with hepatoblastoma, hyperpigmentation and hypopigmentation, autism, and intellectual disability (ID_2086; figure 2, table, appendix). In 2018, Diets and colleagues20 reported a KDM3B truncating mutation in a girl with acute myeloid leukaemia, mild intellectual disability, and hip dysplasia and a de novo non-synonymous KDM3B mutation in a boy with Hodgkins lymphoma and moderate intellectual disability. KDM3B is highly intolerant to both protein-truncating variants (pLI=1·00) and non-synonymous variation (Z=4·99; the Z score is the deviation of observation from expectation for non-synonymous variants). Taken together, these data provide strong evidence that KDM3B is a childhood cancer predisposition gene.

disability. KDM3B is highly intolerant to both protein-truncating variants (pLI=1·00) and non-synonymous variation (Z=4·99; the Z score is the deviation of observation from expectation for non-synonymous variants). Taken together, these data provide strong evidence that KDM3B is a childhood cancer predisposition gene. These new discoveries bring the number of constitutionally mutated genes confirmed as Wilms tumour predisposition genes to 21. We estimate that, together, these constitutional events contribute to about 10% of unselected Wilms tumour (figure 3). Four contributors—WT1, TRIM28, REST, and 11p15 epimutations and uniparental disomy that result in biallelic IGF2 expression—each account for about 2%.2, 4, 5 The remaining 17 are very rare and, together, probably account for no more than 2% of unselected Wilms tumours.2, 3, 4, 6, 7, 8, 9, 10, 11 Functional enrichment analysis highlighted nucleic acid metabolism, chromosome organisation, chromatin or histone modification, and negative regulation of cellular processes as important pathways underlying Wilms tumour predisposition (appendix).Figure 3 Contribution of constitutional mutations to unselected and familial Wilms tumour (A) About 10% of unselected Wilms tumours are due to constitutional mutations in one of 21 genes (pink). (B) A third of familial Wilms tumours are explicable by known Wilms tumour predisposition factors (pink) and two thirds are of unknown cause (blue).

These new discoveries bring the number of constitutionally mutated genes confirmed as Wilms tumour predisposition genes to 21. We estimate that, together, these constitutional events contribute to about 10% of unselected Wilms tumour (figure 3). Four contributors—WT1, TRIM28, REST, and 11p15 epimutations and uniparental disomy that result in biallelic IGF2 expression—each account for about 2%.2, 4, 5 The remaining 17 are very rare and, together, probably account for no more than 2% of unselected Wilms tumours.2, 3, 4, 6, 7, 8, 9, 10, 11 Functional enrichment analysis highlighted nucleic acid metabolism, chromosome organisation, chromatin or histone modification, and negative regulation of cellular processes as important pathways underlying Wilms tumour predisposition (appendix).Figure 3 Contribution of constitutional mutations to unselected and familial Wilms tumour (A) About 10% of unselected Wilms tumours are due to constitutional mutations in one of 21 genes (pink). (B) A third of familial Wilms tumours are explicable by known Wilms tumour predisposition factors (pink) and two thirds are of unknown cause (blue). We have investigated 65 families with two or more cases of Wilms tumour over the last 20 years, including the 49 familial Wilms tumour pedigrees in this study (appendix). In two families, we found a constitutional predisposing mutation in one individual with Wilms tumour, but not their affected relative. We have identified causative constitutional mutations in 22 (35%) of the remaining 63 families (figure 3). The most common of which were mutations in REST (five [8%] of 63), TRIM28 (five [8%] of 63), and WT1 (four [6%] of 63). CTR9 mutations were present in three families and H19 hypermethylation was found in two families. Biallelic BRCA2 mutations, biallelic NYNRIN mutations, and a CDC73 mutation were found in one family each. We identified the CDC73 mutation through this present study (table, appendix). CDC73 is an established Wilms tumour predisposition gene but, to our knowledge, has not previously been associated with familial Wilms tumour. We did not find any cause in two thirds of the families (41 [65%] of 63).

ound in one family each. We identified the CDC73 mutation through this present study (table, appendix). CDC73 is an established Wilms tumour predisposition gene but, to our knowledge, has not previously been associated with familial Wilms tumour. We did not find any cause in two thirds of the families (41 [65%] of 63). Finally, we assessed the overlap between the 21 Wilms tumour predisposition genes and 20 somatically mutated Wilms tumour driver genes (figure 4). Only four genes—WT1, IGF2, TP53, and DICER1—promoted Wilms tumour oncogenesis in both contexts, and all four were also somatically altered in other cancers. A further five constitutionally mutated genes—PIK3CA, FBXW7, ASXL1, BRCA2, and CDC73—were somatically mutated in other cancer types but have not been proven to be somatic drivers in Wilms tumour. The remaining 12 Wilms tumour predisposition genes are not known to be somatically mutated cancer drivers.Figure 4 Overlap of constitutionally and somatically mutated Wilms tumour genes

ASXL1, BRCA2, and CDC73—were somatically mutated in other cancer types but have not been proven to be somatic drivers in Wilms tumour. The remaining 12 Wilms tumour predisposition genes are not known to be somatically mutated cancer drivers.Figure 4 Overlap of constitutionally and somatically mutated Wilms tumour genes Discussion To our knowledge, this is the largest exome sequencing study of individuals with Wilms tumour to date, including 890 affected individuals. Our analyses found three autosomal dominant Wilms tumour predisposition genes—TRIM28, FBXW7, and KDM3B—and one autosomal recessive Wilms tumour predisposition gene, NYNRIN. Constitutional TRIM28 mutations join constitutional WT1 and REST mutations as a relatively common contributor to Wilms tumour predisposition, accounting for about 8% of familial Wilms tumour and about 2% of unselected Wilms tumour.2, 4 We found a strong association between TRIM28 mutations and epithelial Wilms tumour, with most individuals with a TRIM28 mutation having Wilms tumour of predominantly epithelial histology. This suggests that TRIM28 mutations make a sizeable contribution to epithelial Wilms tumour, and we recommend that all children with this rare favourable subtype of Wilms tumour should be offered TRIM28 gene testing.

with most individuals with a TRIM28 mutation having Wilms tumour of predominantly epithelial histology. This suggests that TRIM28 mutations make a sizeable contribution to epithelial Wilms tumour, and we recommend that all children with this rare favourable subtype of Wilms tumour should be offered TRIM28 gene testing. Familial Wilms tumour pedigrees due to TRIM28 mutations showed incomplete penetrance and a strong parent-of-origin effect, because all inherited mutations were maternally transmitted. TRIM28 is not imprinted, but it is located close to PEG3, which is imprinted and paternally expressed.21 At the organism level, PEG3 promotes growth but at the cellular level, it is a putative tumour suppressor.21 A possible explanation for the maternal bias of TRIM28 mutations is that somatic inactivation of the paternal wild-type TRIM28 allele, through mitotic recombination, also leads to loss of the paternally expressed PEG3, thus promoting tumourigenesis. There is precedence for this model in cancer predisposition: SDHD mutations predispose to phaeochromocytoma almost exclusively when inherited paternally.22 The combination of somatic loss of the maternal wild-type SDHD allele and maternally expressed growth inhibitory genes at the IGF1–H19 imprinting region has been proposed as the explanation for this pattern.22 However, a factor against this model for TRIM28 is that the loss of heterozygosity in the tumour from ID_506_01 did not appear to include PEG3.18 Given the prevalence of TRIM28 mutations in Wilms tumour, confirmation of the parent-of-origin bias we observed should be possible in the near future. If this parent-of-origin effect is supported, DNA methylation analyses at the PEG3 imprinting control region and loss of heterozygosity and PEG3 expression analyses in tumours from individuals with TRIM28 mutations might help to provide a mechanistic explanation. From a clinical perspective, establishing if the penetrance of TRIM28 mutations is influenced by the parent-of-origin of the mutation is important, because this would have considerable impact on cancer risks and genetic counselling.

from individuals with TRIM28 mutations might help to provide a mechanistic explanation. From a clinical perspective, establishing if the penetrance of TRIM28 mutations is influenced by the parent-of-origin of the mutation is important, because this would have considerable impact on cancer risks and genetic counselling. We provided evidence that constitutional FBXW7 mutations predisposed to Wilms tumour and to other malignancies. Two of the five individuals carrying FBXW7 mutations had a malignancy other than Wilms tumour. ID_0811 developed osteosarcoma as an adult, after having Wilms tumour. ID_7520 had a rhabdoid tumour and did not have Wilms tumour. The assessment of additional individuals with rhabdoid tumour and de-novo FBXW7 mutations would be useful to further support the role of FBXW7 in rhabdoid tumour predisposition. Furthermore, a woman with Hodgkin lymphoma, adult Wilms tumour, early-onset breast cancer, and a constitutional FBXW7 deletion was reported in 2015,23 and a man with renal cell cancer and a constitutional t(3;4)(q21;q31) translocation disrupting FBXW7 was reported in 2009.24 These data suggest that individuals with FBXW7 mutations might be at risk of multiple childhood and adult cancers and will require ongoing close monitoring. Notably, we believe that the in-frame FBXW7 variant reported25 in an individual with Wilms tumour is not pathogenic because it is not rare and the child also had a pathogenic WT1 mutation.

ndividuals with FBXW7 mutations might be at risk of multiple childhood and adult cancers and will require ongoing close monitoring. Notably, we believe that the in-frame FBXW7 variant reported25 in an individual with Wilms tumour is not pathogenic because it is not rare and the child also had a pathogenic WT1 mutation. KDM3B also appears to be a pleiotropic cancer predisposition gene. The four KDM3B pathogenic mutations reported to date have been associated with four different cancers: Wilms tumour and hepatoblastoma in our study, and acute myeloid leukaemia and Hodgkin lymphoma in the study by Diets and colleagues.20 Large-scale, broad mutation testing of FBXW7 and KDM3B in individuals with cancer will probably be required to establish the full spectrum of associated cancers, because of the rarity of truncating variants and the challenges in interpreting non-synonymous variation in these genes. There are indications that KDM3B and NYNRIN mutations might cause non-malignant phenotypes, particularly intellectual disability. More data on the contribution of these genes to non-malignant conditions will probably become available over the next decade through extensive exome and genome sequencing being done in children with developmental disorders.

and NYNRIN mutations might cause non-malignant phenotypes, particularly intellectual disability. More data on the contribution of these genes to non-malignant conditions will probably become available over the next decade through extensive exome and genome sequencing being done in children with developmental disorders. The four genes we reported here have different functions, and it is unclear why or how they predispose to Wilms tumour. TRIM28 encodes a multidomain protein involved in the regulation of many cellular processes, including transcriptional repression, p53 degradation, pluripotency maintenance, autophagosome formation, epithelial-mesenchyme transition, and the DNA damage response.26 TRIM28 is highly expressed in many cancers, and its inactivation has not been previously associated with oncogenesis. This might explain why inactivating TRIM28 mutations seem to predispose to Wilms tumour alone. The mechanisms underlying this Wilms tumour predisposition are not known, but it is notable that TRIM28 is a major binding partner of AMER1, which is encoded by a gene that is frequently somatically mutated in Wilms tumour.19

This might explain why inactivating TRIM28 mutations seem to predispose to Wilms tumour alone. The mechanisms underlying this Wilms tumour predisposition are not known, but it is notable that TRIM28 is a major binding partner of AMER1, which is encoded by a gene that is frequently somatically mutated in Wilms tumour.19 FBXW7 encodes the substrate recognition component of the E3-ubiquitin ligase SCF complex, which is responsible for recognising and binding phosphorylated substrates and regulating their turnover through proteosome degradation.27 FBXW7 is an established tumour suppressor gene and frequently mutated in many cancers, particularly endometrial and gastrointestinal cancers.27 FBXW7 is not a confirmed somatic driver in Wilms tumour because only one confirmed somatic FBXW7 point mutation has thus far been reported.25 This situation is similar to that of PIK3CA. Constitutional mosaic PIK3CA mutations predispose to Wilms tumour, whereas somatic mutations at the same residue are common in many cancers but have not been reported in Wilms tumour.9 KDM3B encodes a histone H3 demethylase that specifically catalyses the demethylation of H3K9Me1 and H3K9Me2 residues, and is required for normal somatic growth in mice.28 Tumour-suppressive and tumour-promoting KDM3B activities have been proposed in leukaemia, although somatic driver KDM3B mutations have not been reported. Finally, there is very little known about the functions of NYNRIN, though NYN domains are thought to be involved in RNA processing and NYNRIN has been implicated in microRNA–mRNA regulation.29

our-promoting KDM3B activities have been proposed in leukaemia, although somatic driver KDM3B mutations have not been reported. Finally, there is very little known about the functions of NYNRIN, though NYN domains are thought to be involved in RNA processing and NYNRIN has been implicated in microRNA–mRNA regulation.29 The diverse functions of these four new Wilms tumour predisposition genes mirror the broad range of biological processes in which known Wilms tumour predisposition genes operate, as shown by our functional enrichment analysis (appendix). Functional exploration of these genes was beyond the scope of our study, but we hope our results might encourage such assessments, which will probably provide novel insights into oncogenesis and kidney development.

own Wilms tumour predisposition genes operate, as shown by our functional enrichment analysis (appendix). Functional exploration of these genes was beyond the scope of our study, but we hope our results might encourage such assessments, which will probably provide novel insights into oncogenesis and kidney development. Our analyses were designed to identify tumour suppressor genes in which constitutional truncating mutations predisposed to cancer, but were not designed to identify other mechanisms of cancer predisposition. For example, it is very possible that non-truncating coding variation might be contributing to familial and non-familial Wilms tumour, and future analyses to investigate this would be worthwhile. Non-coding, epigenetic, and mosaic abnormalities are all known to be relevant to Wilms tumour predisposition but were not investigated in our study. Notably, none of the known Wilms tumour predisposition genes are within the regions on chromosomes 2p24, 11q14, 5q14, 22q12, and Xp22 identified in a genome-wide association study30 of Wilms tumour, and the causal mechanisms underlying the associations in that study are unknown. Additionally, the mutations at 17q21 responsible for FWT1-linked families have not yet been discovered.2

within the regions on chromosomes 2p24, 11q14, 5q14, 22q12, and Xp22 identified in a genome-wide association study30 of Wilms tumour, and the causal mechanisms underlying the associations in that study are unknown. Additionally, the mutations at 17q21 responsible for FWT1-linked families have not yet been discovered.2 Genetic predisposition to Wilms tumour exhibits remarkable heterogeneity, and this is particularly noteworthy because childhood cancers are generally assumed to be aetiologically simpler than adult cancers. Furthermore, our study provides strong evidence that further genetic, genomic, or epigenetic Wilms tumour predisposition factors exist, because only a third of the familial Wilms tumour pedigrees we investigated have been explained. Any further familial Wilms tumour genes discovered will be highly likely to contribute also to non-familial Wilms tumour. Our study reveals new insights into the complexity, mechanisms, and clinical implications of Wilms tumour predisposition. Although our understanding of the genetic landscape of Wilms tumour predisposition is still far from complete, the available knowledge has considerable scientific and clinical use. Given the extensive heterogeneity and the absence of family history or additional clinical features in many individuals with a mutation, we believe routine genetic testing in all individuals with Wilms tumour would be scientifically and clinically valuable. Supplementary Material Supplementary appendix

Our study reveals new insights into the complexity, mechanisms, and clinical implications of Wilms tumour predisposition. Although our understanding of the genetic landscape of Wilms tumour predisposition is still far from complete, the available knowledge has considerable scientific and clinical use. Given the extensive heterogeneity and the absence of family history or additional clinical features in many individuals with a mutation, we believe routine genetic testing in all individuals with Wilms tumour would be scientifically and clinically valuable. Supplementary Material Supplementary appendix Acknowledgments We thank the families for their participation and the many doctors, nurses, and counsellors who recruited them to the FACT study. The FACT collaborators are listed in the appendix. This study was funded by the Wellcome Trust (100210/Z/12/Z). We acknowledge support of the National Institute for Health Research Clinical Research Network (NIHR CRN), the Children's Cancer and Leukaemia Group (CCLG), and the Royal Marsden-ICR NIHR Biomedical Research Centre. We thank Jessie Bull for assistance in FACT recruitment and Ann Strydom for assistance in preparing the manuscript. Contributors SM, SY, ER, and NR designed the study. SM, EP-P, SS, and SH did the molecular studies. SM, SY, EH, AE, MC, and ER handled data management, data analyses, or both. JA, SB, TC, RFa, RFu, AG, RG, JH, SL, FM, JN, MR, JS, DW, and DY contributed to the sample and data collection coordinated by AZ and MW-P. NR, SM, and SY wrote the manuscript with input from the other authors.

and SH did the molecular studies. SM, SY, EH, AE, MC, and ER handled data management, data analyses, or both. JA, SB, TC, RFa, RFu, AG, RG, JH, SL, FM, JN, MR, JS, DW, and DY contributed to the sample and data collection coordinated by AZ and MW-P. NR, SM, and SY wrote the manuscript with input from the other authors. Declaration of interests NR reports personal fees from AstraZeneca and Genomics, outside the submitted work. ER reports personal fees from Foresite Capital. JA reports personal fees from TC Biopharm and holds founder shares in Autolus Therapeutics. All other authors declare no competing interests.

Introduction Perinatal HIV infection is a major risk factor for child undernutrition, but programmes to prevent mother-to-child transmission have reduced new infant infections by almost half since 2010.1 However, there is an increasing population of HIV-exposed uninfected children,2 who have more undernutrition than HIV-unexposed children. In a birth cohort of 14 110 Zimbabwean infants, HIV-exposed uninfected children had 23% increased odds of stunting and 56% increased odds of wasting at 12 months of age,3 and mean head circumference was lower throughout the first year of life compared with HIV-unexposed infants.4 Infants born to HIV-positive mothers are therefore vulnerable to undernutrition, regardless of their own infection status, and targeted interventions to promote healthy growth are required. Research in context Evidence before this study

Introduction Perinatal HIV infection is a major risk factor for child undernutrition, but programmes to prevent mother-to-child transmission have reduced new infant infections by almost half since 2010.1 However, there is an increasing population of HIV-exposed uninfected children,2 who have more undernutrition than HIV-unexposed children. In a birth cohort of 14 110 Zimbabwean infants, HIV-exposed uninfected children had 23% increased odds of stunting and 56% increased odds of wasting at 12 months of age,3 and mean head circumference was lower throughout the first year of life compared with HIV-unexposed infants.4 Infants born to HIV-positive mothers are therefore vulnerable to undernutrition, regardless of their own infection status, and targeted interventions to promote healthy growth are required. Research in context Evidence before this study In low-income countries, child stunting is highly prevalent, particularly in HIV-exposed uninfected children compared with HIV-unexposed children. We searched PubMed with the following search terms: (stunting OR length OR height OR haemoglobin OR hemoglobin OR anaemia OR anemia) AND (child OR infant) AND (feeding OR WASH OR water OR sanitation OR hygiene) AND (HIV) from database inception up to June 1, 2018 with no language restrictions. An observational study from Tanzania reported less stunting among HIV-exposed children with higher infant and child feeding index scores, and observational data from the Côte d'Ivoire showed that improved complementary feeding at age 6 months was associated with better linear growth and a lower prevalence of stunting during the subsequent 12 months. A non-controlled intervention of nutritional support for non-breastfed, HIV-exposed Haitian children found a reduction in stunting compared with an historical control population at 6 months and 12 months of age. In trial data of maternal nutritional supplementation for breastfeeding mothers, there was no effect on linear growth among HIV-exposed children in South Africa or Malawi. Modest improvements in linear growth of HIV-exposed children have been shown in trials of improved infant feeding among infants breastfed for only short periods (<6 months) or not at all. Provision of lipid-based nutrient supplements instead of breastfeeding from 6 months of age improved linear growth in Zambian HIV-exposed children. In formula-fed, HIV-exposed uninfected children in the USA, Bahamas, and Brazil, more concentrated formula improved weight compared with standard formula. We found one study assessing the effect of nutrition on haemoglobin in HIV-exposed infants: in Zambia, rich fortification of porridge increased haemoglobin concentration and reduced anaemia. We did not identify any studies of WASH or combined WASH and complementary feeding on linear growth of HIV-exposed children. We did not identify any studies investigating the relationship between WASH and haemoglobin concentration or anaemia.

fication of porridge increased haemoglobin concentration and reduced anaemia. We did not identify any studies of WASH or combined WASH and complementary feeding on linear growth of HIV-exposed children. We did not identify any studies investigating the relationship between WASH and haemoglobin concentration or anaemia. Added value of this study To our knowledge, this is the first randomised trial to investigate the effect of improved water, sanitation, and hygiene (WASH) and improved infant and young child feeding (IYCF)—independently and in combination—in HIV-exposed children. This study was done in a rural African setting with high antenatal HIV prevalence, and high rates of child stunting and anaemia. IYCF and WASH interventions were informed by extensive formative research, were culturally appropriate, and led to significant behaviour change. Consistent with previous studies, we found that complementary feeding improved linear growth and haemoglobin concentration in HIV-exposed children, reducing stunting by 20% and anaemia by 50%. We found no effect on linear growth or haemoglobin concentration with the WASH intervention. Implications of all the available evidence

To our knowledge, this is the first randomised trial to investigate the effect of improved water, sanitation, and hygiene (WASH) and improved infant and young child feeding (IYCF)—independently and in combination—in HIV-exposed children. This study was done in a rural African setting with high antenatal HIV prevalence, and high rates of child stunting and anaemia. IYCF and WASH interventions were informed by extensive formative research, were culturally appropriate, and led to significant behaviour change. Consistent with previous studies, we found that complementary feeding improved linear growth and haemoglobin concentration in HIV-exposed children, reducing stunting by 20% and anaemia by 50%. We found no effect on linear growth or haemoglobin concentration with the WASH intervention. Implications of all the available evidence The findings of this trial, supported by previous studies in low-income and middle-income countries, show that IYCF modestly improves linear growth and haemoglobin concentrations in HIV-exposed children, and would lead to substantial reductions in stunting and anaemia if delivered at scale. Because HIV-exposed children are particularly vulnerable to undernutrition, populations with a high maternal HIV prevalence may particularly benefit from complementary feeding interventions. However, combining complementary feeding with the elementary household-level WASH interventions commonly implemented in rural areas of low-income and middle-income countries (ie, pit latrines, hand-washing stations not connected to a water source, and point-of-use chlorination of drinking water with monthly behaviour-change promotion) provides no additional benefit compared with IYCF alone, indicating that more effective WASH interventions are required.

w-income and middle-income countries (ie, pit latrines, hand-washing stations not connected to a water source, and point-of-use chlorination of drinking water with monthly behaviour-change promotion) provides no additional benefit compared with IYCF alone, indicating that more effective WASH interventions are required. Stunting, the most prevalent form of undernutrition, is associated with increased child mortality5 and reduced school attainment, and perpetuates an intergenerational cycle of inequity.6 Anaemia often co-exists with stunting, is similarly intractable,7 and is another major cause of impaired neurodevelopment.8 Since optimising infant and young child feeding (IYCF) only modestly improves linear growth,9 there is an increasing recognition that multisectoral approaches are required to tackle stunting. Improving water, sanitation, and hygiene (WASH) might affect growth and anaemia by reducing diarrhoeal disease and preventing environmental enteric dysfunction, a subclinical inflammatory disorder of the small intestine that is highly prevalent among children living in poverty.10 The Sanitation Hygiene Infant Nutrition Efficacy (SHINE) trial was designed to test the independent and combined effects of improved IYCF and improved WASH, on both stunting and anaemia in an area of high antenatal HIV prevalence in rural Zimbabwe. We have previously reported trial results of children born to HIV-negative women;11 here, we report the effect of IYCF and WASH on stunting and anaemia among HIV-exposed children.

ined effects of improved IYCF and improved WASH, on both stunting and anaemia in an area of high antenatal HIV prevalence in rural Zimbabwe. We have previously reported trial results of children born to HIV-negative women;11 here, we report the effect of IYCF and WASH on stunting and anaemia among HIV-exposed children. Methods Study design and participants The SHINE trial design has been reported previously;10, 11 the protocol and statistical analysis plan are available online. Briefly, SHINE was a cluster-randomised community-based 2 × 2 factorial trial done in two contiguous rural districts in Zimbabwe with 15% antenatal HIV prevalence. Clusters were defined as the catchment area of 1–4 village health workers employed by the Ministry of Health and Child Care. Village health workers did prospective pregnancy surveillance and established date of last menstrual period among pregnant women, and referred pregnant women to SHINE research nurses for trial enrolment. Women were eligible if they permanently resided in a study cluster and were confirmed pregnant. Over the recruitment period, the cutoff of gestational age for recruitment eligibility was gradually liberalised to maximise recruitment (appendix). The Medical Research Council of Zimbabwe and the Institutional Review Board of the Johns Hopkins Bloomberg School of Public Health approved the study protocol. All participants provided written informed consent.

of gestational age for recruitment eligibility was gradually liberalised to maximise recruitment (appendix). The Medical Research Council of Zimbabwe and the Institutional Review Board of the Johns Hopkins Bloomberg School of Public Health approved the study protocol. All participants provided written informed consent. Randomisation and masking Clusters were allocated (1:1:1:1) to one of four treatment groups: standard of care, IYCF, WASH, or IYCF plus WASH at a public event. A highly constrained randomisation technique achieved balance across groups for 14 variables related to geography, demography, water access, and sanitation coverage (appendix).12 Masking of participants and fieldworkers was not possible, but investigators analysing the data were masked to group allocation.

ublic event. A highly constrained randomisation technique achieved balance across groups for 14 variables related to geography, demography, water access, and sanitation coverage (appendix).12 Masking of participants and fieldworkers was not possible, but investigators analysing the data were masked to group allocation. Procedures Interventions were informed by extensive formative research and piloting.10 All women were scheduled to receive 15 modules delivered by group-specific village health workers (VHWs), with behaviour-change messages and interactive tools between enrolment and 12 months postnatal (approximately one visit per month); other family members were encouraged to participate. At each visit, previous information was reviewed before introducing new information to create a sequenced integrated longitudinal intervention. Between 13 and 17 months, VHWs visited monthly, providing routine care and delivering intervention supplies; during these visits VHWs encouraged participants to practise relevant behaviours, although structured modules were not implemented (see key messages and supplies in the appendix, and lesson plans and interactive tools online).

17 months, VHWs visited monthly, providing routine care and delivering intervention supplies; during these visits VHWs encouraged participants to practise relevant behaviours, although structured modules were not implemented (see key messages and supplies in the appendix, and lesson plans and interactive tools online). Standard of care messages comprised promotion of exclusive breastfeeding to 6 months, uptake of antenatal and neonatal care, prevention of mother-to-child HIV transmission, immunisations, family planning, and standard IYCF information based on WHO recommend-ations. Groups with the IYCF component received all standard of care messages plus information about the importance of nutrition for infant health, growth, and development; feeding nutrient-dense food and a 20 g small-quantity lipid-based nutrient supplement (SQ-LNS; Nutriset, Malaumay, France) daily from age 6 months to 18 months; processing locally available foods to facilitate mastication and swallowing; feeding during illness; and dietary diversity. The IYCF modules therefore addressed specific contextual barriers through a sequential longitudinal intervention based on successive messages and reinforcement. VHWs also made monthly deliveries of 30 20 g sachets of small-quantity lipid-based nutrient supplement from infant age 6 months through to 18 months. The WASH component included all standard of care messages plus information about safe disposal of faeces; hand-washing with soap after faecal contact and before preparing food, eating food or feeding children; protection of infants from geophagia and animal faeces ingestion; chlorination of drinking water; and hygienic preparation of complementary food. Additionally, a ventilated improved pit latrine was provided within 6 weeks of enrolment; two hand-washing stations, plastic mat and play yard (North States, Minneapolis, MN, USA), and monthly delivery of soap and chlorine (WaterGuard, Nelspot, Zimbabwe) were provided. A latrine was constructed in the non-WASH groups following trial completion.

improved pit latrine was provided within 6 weeks of enrolment; two hand-washing stations, plastic mat and play yard (North States, Minneapolis, MN, USA), and monthly delivery of soap and chlorine (WaterGuard, Nelspot, Zimbabwe) were provided. A latrine was constructed in the non-WASH groups following trial completion. Research nurses made home visits at baseline (approximately 2 weeks after consent), 32 weeks' gestation, and at 1 month, 3 months, 6 months, 12 months, and 18 months post partum to assess maternal and household characteristics and trial outcomes. At baseline, mothers had height, weight, and mid-upper arm circumference measured, and were tested for haemoglobin concentrations (Hemocue, Ängelholm, Sweden), Schistosoma haematobium infection (by urinary microscopy), and HIV. HIV-positive women were encouraged to seek immediate antenatal care to prevent mother-to-child transmission. Other maternal and household characteristics were assessed, including dietary diversity, food insecurity, household wealth, and maternal capabilities.13

ma haematobium infection (by urinary microscopy), and HIV. HIV-positive women were encouraged to seek immediate antenatal care to prevent mother-to-child transmission. Other maternal and household characteristics were assessed, including dietary diversity, food insecurity, household wealth, and maternal capabilities.13 Infant birth date, weight, and delivery details were transcribed from health facility records. The trial provided Tanita BD-590 infant scales (Weigh & Measure, Olney, MD, USA) to all health institutions in the study area and trained facility staff. Gestational age at delivery was calculated from last menstrual period dates. At the 18-month postnatal visit (trial endpoint), mothers and infants were visited anywhere in the country for the intention-to-treat analyses of primary outcomes; however, given the household-based nature of the interventions, intermediate visits were done only when the mother was available in the household where she consented. At 18 months postnatal, infant point-of-care haemoglobin concentration was measured (HemoCue, Ängelholm, Sweden). Infant length was calculated as the median of three measurements; weight, head circumference, and mid-upper arm circumference were also measured (appendix). Infant diarrhoea (three or more loose or watery stools in 24 h), dysentery (stool with blood or mucus), and acute respiratory infection (fast or difficult breathing) were assessed by 7 day maternal recall at postnatal visits. Infants with acute malnutrition or illness were referred to clinics.

also measured (appendix). Infant diarrhoea (three or more loose or watery stools in 24 h), dysentery (stool with blood or mucus), and acute respiratory infection (fast or difficult breathing) were assessed by 7 day maternal recall at postnatal visits. Infants with acute malnutrition or illness were referred to clinics. Adverse events and serious adverse events were ascertained by research nurses during visits, and by village health workers during intervention delivery contacts, and reported to a senior research nurse who collected details. Events were reviewed by the study physician (AJP) to determine relatedness to trial interventions before reporting to the responsible institutional review boards. An independent data safety and monitoring board comprising two physicians from Zimbabwe and a statistician from the UK reviewed interim adverse event data.

. Events were reviewed by the study physician (AJP) to determine relatedness to trial interventions before reporting to the responsible institutional review boards. An independent data safety and monitoring board comprising two physicians from Zimbabwe and a statistician from the UK reviewed interim adverse event data. Mothers were tested for HIV status at baseline with a rapid test algorithm (Determine HIV 1/2 test [Alere International Limited, Ballybrit, Ireland], followed by INSTI HIV 1/2 test [bioLytical Laboratories Inc., Richmond, BC, Canada] if positive). HIV-positive women had CD4 counts measured (Pima Analyser [Alere International Limited, Ballybrit, Ireland]) and were referred to local clinics. Viral load was not measured in the trial. National guidelines for prevention of mother-to-child HIV transmission changed from WHO Option B (maternal antiretrovial therapy [ART] from 14 gestational weeks until the end of breastfeeding) to Option B+ (lifelong ART for all pregnant and breastfeeding women) in November, 2013. Women were encouraged to initiate co-trimoxazole and ART, to exclusively breastfeed, and to attend clinic at 6 weeks post partum for early infant diagnosis and initiation of infant co-trimoxazole. Women testing HIV-negative at baseline were retested at 32 gestational weeks and 18 months post partum to detect sero-conversion.

were encouraged to initiate co-trimoxazole and ART, to exclusively breastfeed, and to attend clinic at 6 weeks post partum for early infant diagnosis and initiation of infant co-trimoxazole. Women testing HIV-negative at baseline were retested at 32 gestational weeks and 18 months post partum to detect sero-conversion. HIV-positive mothers were invited to enrol in a substudy, in which infant blood was collected at 1 month, 3 months, 6 months, 12 months, and 18 months and tested for HIV; infants of mothers declining substudy enrolment were only tested at 18 months. Children were classified as HIV-positive or HIV-exposed uninfected based on results at 18 months, or their last available test. Children not tested at 18 months owing to caregiver refusal, defaulted visits, or loss to follow-up were classified as HIV-unknown. Inconclusive or discordant results were retested to confirm status; if no further samples were available or repeat testing was inconclusive, children were classified as HIV-unknown. Before 18 months of age, HIV was diagnosed by DNA PCR on dried blood-spot samples or RNA PCR on plasma; and after 18 months, by PCR or rapid test algorithm, depending on samples provided. All HIV-positive children were referred for ART initiation.

epeat testing was inconclusive, children were classified as HIV-unknown. Before 18 months of age, HIV was diagnosed by DNA PCR on dried blood-spot samples or RNA PCR on plasma; and after 18 months, by PCR or rapid test algorithm, depending on samples provided. All HIV-positive children were referred for ART initiation. Outcomes The coprimary outcomes were length for age Z score and haemoglobin concentration in infants at age 18 months (allowable age range 76–130 weeks). Secondary outcomes were stunting (length for age Z score <–2), severe stunting (length for age Z score <–3), anaemia (haemoglobin <105 g/L), severe anaemia (<70 g/L); weight for age Z score, underweight (weight for age Z score <–2), weight for length Z score, wasting (weight for length Z score <–2), mid-upper arm circumference for age Z score and head circumference for age Z score at 18 months; 7-day maternal recall of diarrhoea, dysentery, and acute respiratory infection at 12 months and 18 months; and all-cause mortality up to 18 months. Intervention uptake was assessed at all visits and reported here for the 12-month visit. Statistical analysis Sample size calculation was done for HIV-unexposed infants.11 No specific sample size calculation for outcomes among HIV-exposed infants was done. All analyses were done on an intention-to-treat basis at the child level.

Outcomes The coprimary outcomes were length for age Z score and haemoglobin concentration in infants at age 18 months (allowable age range 76–130 weeks). Secondary outcomes were stunting (length for age Z score <–2), severe stunting (length for age Z score <–3), anaemia (haemoglobin <105 g/L), severe anaemia (<70 g/L); weight for age Z score, underweight (weight for age Z score <–2), weight for length Z score, wasting (weight for length Z score <–2), mid-upper arm circumference for age Z score and head circumference for age Z score at 18 months; 7-day maternal recall of diarrhoea, dysentery, and acute respiratory infection at 12 months and 18 months; and all-cause mortality up to 18 months. Intervention uptake was assessed at all visits and reported here for the 12-month visit. Statistical analysis Sample size calculation was done for HIV-unexposed infants.11 No specific sample size calculation for outcomes among HIV-exposed infants was done. All analyses were done on an intention-to-treat basis at the child level. For primary analyses, we used generalised estimating equations that accounted for within-cluster correlation and contained two dummy variables representing the main effect of the IYCF intervention (the two IYCF-containing groups compared to the two groups without IYCF) and the WASH intervention (the two WASH-containing groups compared to the two groups without WASH), unadjusted for other covariates, with an exchangeable working correlation structure.11 Although the study was not powered to detect a statistical interaction between the IYCF and WASH interventions, we estimated these interactions for each outcome. When the interaction was significant (ie, p<0·05 according to the Wald test) or had a sizeable point estimate (ie, RR >2 or <0·5 when comparing ratio-of-ratios, or difference-of-differences >0·25 SDs when comparing continuous outcomes), results are based on a regression model with three dummy variables to represent IYCF, WASH and IYCF plus WASH compared to standard of care instead of the model of two terms. In adjusted analyses we controlled for prespecified baseline covariates, which were initially assessed in bivariate analyses to identify those with an important association with the outcome (ie, p<0·2 or RR >2·0 or <0·5 for dichotomous outcomes, and p<0·2 or difference >0·25 SDs for continuous outcomes). Selected covariates were entered in a multivariable regression model; a forward stepwise selection procedure was implemented with p<0·2 to enter. A log-binomial specification was used to facilitate estimation of relative risks (RR). Depending on the analysis, other methods for comparing groups while accounting for within-cluster correlation included multinomial and ordinal regression models with robust variance estimation, and Somers' D for medians.

·2 to enter. A log-binomial specification was used to facilitate estimation of relative risks (RR). Depending on the analysis, other methods for comparing groups while accounting for within-cluster correlation included multinomial and ordinal regression models with robust variance estimation, and Somers' D for medians. In a per-protocol analysis, we examined the effect of the interventions when behaviour-change modules were delivered at high fidelity (which was predefined for the IYCF plus WASH group as receiving all ten core modules and for the other study groups as receiving all modules scheduled at the same timepoints when IYCF plus WASH core modules were delivered). A prespecified subgroup analysis of primary outcomes by infant sex was planned if the interaction terms were significant (p<0·05). A sensitivity analysis excluded children testing HIV-positive or HIV-unknown at 18 months. We used Stata (version 14) for all analyses. The study is registered with ClinicalTrials.gov, number NCT01824940. Role of the funding source The funders of the study had no role in study design, data collection, data analysis, data interpretation, or writing of the report. The corresponding author had full access to all the data in the study and had final responsibility for the decision to submit for publication.

We used Stata (version 14) for all analyses. The study is registered with ClinicalTrials.gov, number NCT01824940. Role of the funding source The funders of the study had no role in study design, data collection, data analysis, data interpretation, or writing of the report. The corresponding author had full access to all the data in the study and had final responsibility for the decision to submit for publication. Results Between Nov 22, 2012, and March 27, 2015, 5280 pregnant women were enrolled from 211 clusters at a median gestational age of 12 (IQR 9–16) weeks (figure). Among 4843 livebirths, 738 infants were born to 726 mothers testing HIV-positive during pregnancy and are included in this analysis. During the postnatal period, 51 (7%) of 738 infants died, of whom three were HIV-positive, two HIV-exposed uninfected, and 46 HIV-unknown.Figure Trial profile

2 (IQR 9–16) weeks (figure). Among 4843 livebirths, 738 infants were born to 726 mothers testing HIV-positive during pregnancy and are included in this analysis. During the postnatal period, 51 (7%) of 738 infants died, of whom three were HIV-positive, two HIV-exposed uninfected, and 46 HIV-unknown.Figure Trial profile SOC=standard of care. IYCF=infant and young child feeding. WASH=water, sanitation, and hygiene. *212 clusters were randomly assigned, 53 in each of the four trial groups. After randomisation, one cluster was excluded because it was in an urban area, one was excluded because the village health worker covering it mainly had clients outside the study area, and two more were merged on the basis of subsequent data for village health worker coverage. Three new cluster designations were created because of anomalies in the original mapping. For two of these cluster, the trial group was clear; the third contained areas that were in two trial groups, and was assigned to the under-represented group, resulting in 53 clusters in each group. All these changes occurred before enrolment began. When enrolment was completed, however, no women were enrolled in one cluster in the SOC group and thus 211 clusters were available for analysis.

reas that were in two trial groups, and was assigned to the under-represented group, resulting in 53 clusters in each group. All these changes occurred before enrolment began. When enrolment was completed, however, no women were enrolled in one cluster in the SOC group and thus 211 clusters were available for analysis. Among households of HIV-positive women at baseline, one-third had a latrine, two-thirds had some solar powered electricity, and around half of all household members practised open defecation (table 1). Water access was poor. Most mothers were married and had completed on average 9 years of schooling but few had employment. Nutritional status was generally good; no women had mid-upper arm circumference less than 19 cm. Mean CD4 count at baseline was 473 cells/μL (SD 221); most mothers received ART and just over half received co-trimoxazole during pregnancy (table 1). There were some minor baseline imbalances between groups. Women in the standard-of-care group were slightly poorer, with a lower wealth index and higher proportion of homes with no electricity and unimproved floors. More households in the IYCF plus WASH and WASH groups had a hand-washing station at baseline. Women in the WASH and IYCF plus WASH groups had slightly lower mean CD4 counts; mothers in the WASH group were less likely to be taking ART but more likely to be taking co-trimoxazole. Infant birth characteristics were similar across treatment groups. The majority were born by vaginal delivery and had institutional deliveries. Mean birthweight was 2·99 (SD 0·50) kg.Table 1 Maternal, household, and infant baseline characteristics of HIV-positive mothers and their liveborn infants

aking co-trimoxazole. Infant birth characteristics were similar across treatment groups. The majority were born by vaginal delivery and had institutional deliveries. Mean birthweight was 2·99 (SD 0·50) kg.Table 1 Maternal, household, and infant baseline characteristics of HIV-positive mothers and their liveborn infants Standard of care IYCF WASH IYCF plus WASH Mothers 166 155 200 205 Infants 166 158 205 209 Mothers completing baseline visit 164 151 199 203 Household characteristics Median number of occupants (IQR) 4 (3–5) 4 (3–6) 5 (3–6) 4 (3–6) Wealth quintile14 Lowest 46/163 (28%) 39/151 (26%) 50/197 (25%) 56/201 (28%) Second 47/163 (29%) 27/151 (18%) 44/197 (22%) 49/201 (24%) Middle 32/163 (20%) 31/151 (21%) 39/197 (20%) 34/201 (17%) Fourth 18/163 (11%) 35/151 (23%) 27/197 (14%) 26/201 (13%) Highest 20/163 (12%) 19/151 (13%) 37/197 (19%) 36/201 (18%) Electricity Power grid 5/163 (3%) 3/151 (2%) 4/197 (2%) 7/201 (4%) Other power source Generator 4/163 (3%) 7/151 (5%) 2/197 (1%) 5/202 (3%) Solar 96/163 (59%) 102/151 (68%) 127/197 (65%) 128/202 (64%) No electricity 63/163 (39%) 42/151 (28%) 68/197 (35%) 69/202 (34%) Sanitation Household members who openly defecate 366/609 (60%) 349/621 (56%) 392/788 (50%) 402/780 (52%) Any latrine at household 46/163 (28%) 56/147 (38%) 82/196 (42%) 74/197 (38%) Improved latrine at household 38/163 (23%) 48/147 (33%) 74/196 (38%) 66/196 (34%) Improved latrine with well-trodden path and not shared 21/157 (13%) 37/144 (26%) 53/193 (28%) 39/188 (21%) Water Main source of household drinking water is improved 101/163 (62%) 88/147 (60%) 114/194 (59%) 119/197 (60%) Treat drinking water to make it safer 20/160 (13%) 20/144 (14%) 20/192 (10%) 24/197 (12%) Median one-way walk time to fetch water (IQR), min 10 (5–20) 10 (3–15) 10 (5–20) 10 (5–20) Mean per-capita water volume collected in past 24 h (SD), L 8·7 (3·7) [n=125] 9·9 (7·6) [n=121] 9·5 (5·9) [n=160] 9·7 (7·6) [n=169] Hygiene Handwashing station at household 4/141 (3%) 8/142 (6%) 32/188 (17%) 24/187 (13%) Handwashing station with water and rubbing agent 0/141 0/138 0/187 1/187 (<1%) Improved floor* 66/162 (41%) 75/150 (50%) 97/194 (50%) 99/196 (51%) Median number of chickens (IQR) 4 (0–9) 4 (0–10) 5 (2–9) 4 (0–8) Livestock observed inside home 64/164 (39%) 45/152 (30%) 71/196 (36%) 63/199 (32%) Faeces observed in yard 58/162 (36%) 42/152 (28%) 62/194 (32%) 47/198 (24%) Diet quality and food security Household meets minimum dietary diversity15 51/137 (37%) 53/130 (41%) 62/174 (36%) 70/171 (41%) Med

(0–9) 4 (0–10) 5 (2–9) 4 (0–8) Livestock observed inside home 64/164 (39%) 45/152 (30%) 71/196 (36%) 63/199 (32%) Faeces observed in yard 58/162 (36%) 42/152 (28%) 62/194 (32%) 47/198 (24%) Diet quality and food security Household meets minimum dietary diversity15 51/137 (37%) 53/130 (41%) 62/174 (36%) 70/171 (41%) Med ian coping strategies Index (IQR)16 3 (0–10) 2 (0–13) 2 (0–9·5) 3 (0–11) Maternal characteristics Mean age (SD), years 29·7 (6·0) 29·0 (6·8) 28·4 (4·8) 29·6 (5·7) Mean height (SD), cm 161·1 (7·2) 160·0 (5·8) 160·5 (6·9) 159·4 (6·5) Mean mid-upper-arm circumference (SD), cm 26·4 (3·4) 26·1 (3·0) 26·4 (2·1) 26·1 (2·6) Positive microscopy for Schistosoma haematobium 16/154 (10%) 10/148 (7%) 24/185 (13%) 21/192 (11%) Mean years of completed schooling (SD) 9·4 (1·8) 9·0 (2·2) 8·9 (2·3) 9·1 (2·4) Median parity (IQR) 2 (1–3) [n=113] 2 (1–3) [n=110] 2 (1–3) [n=133] 2 (1–3) [n=160] Married 148/156 (95%) 140/149 (94%) 178/188 (95%) 177/189 (94%) Employed 17/160 (11%) 16/151 (11%) 22/197 (11%) 12/202 (6%) Religion Apostolic 77/158 (49%) 67/149 (45%) 88/189 (47%) 98/193 (51%) Other Christian religions 66/158 (42%) 67/149 (45%) 77/189 (41%) 78/193 (40%) Other non-Christian religions 15/158 (10%) 15/149 (10%) 24/189 (13%) 17/193 (9%) HIV disease severity and treatment Mean CD4 count in pregnancy (SD), cells per μL† 503 (215) [n=129] 496 (191) [n=136] 444 (184) [n=163] 464 (186) [n=175] Antiretroviral therapy during pregnancy‡ 139/166 (84%) 130/155 (84%) 151/200 (76%) 167/205 (82%) Co-trimoxazole prophylaxis during pregnancy§ 84/166 (51%) 84/155 (54%) 123/200 (62%) 111/205 (54%) Infant characteristics Sex Female 80/165 (48%) 77/156 (49%) 97/205 (47%) 113/207 (55%) Male 85/165 (52%) 79/156 (51%) 108/205 (53%) 94/207 (45%) Mean birthweight (SD), kg 3·00 (0·50) 2·99 (0·41) 3·03 (0·67) 2·99 (0.59) Birthweight <2500 g 20/166 (12%) 18/158 (11%) 23/205 (11%) 23/209 (11%) Institutional delivery 119/144 (83%) 119/142 (84%) 148/176 (84%) 149/176 (85%) Vaginal delivery 139/147 (95%) 131/141 (93%) 165/179 (92%) 166/183 (91%) IYCF=infant and young child feeding.

ht (SD), kg 3·00 (0·50) 2·99 (0·41) 3·03 (0·67) 2·99 (0.59) Birthweight <2500 g 20/166 (12%) 18/158 (11%) 23/205 (11%) 23/209 (11%) Institutional delivery 119/144 (83%) 119/142 (84%) 148/176 (84%) 149/176 (85%) Vaginal delivery 139/147 (95%) 131/141 (93%) 165/179 (92%) 166/183 (91%) IYCF=infant and young child feeding. WASH=water, sanitation, and hygiene. Baseline for mothers was 2 weeks after consent (approximately 14 weeks gestation). Baseline for infants was at birth. Values are n (%), unless stated. * Improved floor defined as concrete, brick, cement, or tile. Unimproved floor defined as mud, earth, sand, or dung. † CD4 count at baseline visit, or at 32 gestational week visit if no baseline result. ‡ Includes any exposure to antiretroviral therapy during pregnancy. § Includes any exposure to co-trimoxazole during pregnancy.

* Improved floor defined as concrete, brick, cement, or tile. Unimproved floor defined as mud, earth, sand, or dung. † CD4 count at baseline visit, or at 32 gestational week visit if no baseline result. ‡ Includes any exposure to antiretroviral therapy during pregnancy. § Includes any exposure to co-trimoxazole during pregnancy. Fidelity of intervention delivery by the trial was high (table 2). Among WASH households, almost all received latrines and handwashing stations, the vast majority received a play mat and yard, and receipt of soap and chlorine was high. Among IYCF households, receipt of SQ-LNS was similarly high. Across all groups, VHWs completed the vast majority of planned intervention visits, although module delivery was slightly higher in the IYCF groups than in the non-IYCF groups (table 2). Intervention uptake was assessed by participant behaviours at the 12-month postnatal visit, when three-quarters of women were in their primary home and available for the visit (table 2). Women assessed at 12 months were, on average, 3 years older, with similar parity, more likely to be married, and less likely to be employed at baseline compared with women not assessed at 12 months. Baseline indicators of diet, water, sanitation, and hygiene were similar between women who were and were not assessed for intervention uptake at 12 months (appendix).Table 2 Intervention delivery and participant uptake by treatment group

s likely to be employed at baseline compared with women not assessed at 12 months. Baseline indicators of diet, water, sanitation, and hygiene were similar between women who were and were not assessed for intervention uptake at 12 months (appendix).Table 2 Intervention delivery and participant uptake by treatment group Data source Standard of care IYCF WASH IYCF plus WASH Combined WASH* Non-WASH* p value Combined IYCF† Non-IYCF† p value Intervention delivery at baseline Children with 18-month outcomes (on whom inferences are based), n Trial logs 147 147 184 190 374 294 .. 337 331 .. WASH supplies SHINE-installed ventilated improved pit latrine Trial logs NA NA 181/184 (98%) 188/190 (99%) 369/374 (99%) NA .. NA NA .. Two handwashing stations (ie, Tippy Taps) delivered Trial logs NA NA 184/184 (100%) 190/190 (100%) 374/374 (100%) NA .. NA NA .. Baby mat delivered Trial logs NA NA 174/184 (95%) 183/190 (96%) 358/374 (96%) NA .. NA NA .. Play yard delivered Trial logs NA NA 171/184 (92%) 180/190 (95%) 351/374 (94%) NA .. NA NA .. Median liquid soap deliveries (IQR)‡ Trial logs NA NA 19 (18–20) 20 (18–20) 20 (18–20) NA .. NA NA .. Received at least 16 (80% of expected) soap deliveries Trial logs NA NA 145/184 (79%) 159/190 (84%) 304/374 (81%) NA .. NA NA .. Median WaterGuard deliveries (IQR)‡ Trial logs NA NA 15 (14–15) 15 (15–15) 15 (14–15) NA .. NA NA .. Received at least 12 (80% of expected) WaterGuard deliveries Trial logs NA NA 150/184 (82%) 159/190 (84%) 309/374 (83%) NA .. NA NA .. IYCF supplies Median SQ-LNS deliveries (IQR) Trial logs NA 13 (12–13) NA 13 (13–13) NA NA .. 13 (12–13) NA .. Received ≥11 (80% of expected) SQ-LNS deliveries Trial logs NA 117/147 (80%) NA 157/190 (83%) NA NA .. 274/337 (81%) NA .. Behaviour change modules Median intervention modules (IQR) VHW report 15 (11–15) 15 (14–15) 15 (13·5–15) 15 (14–15) 15 (14–15) 15 (13–15) 0·58 15 (14–15) 15 (13–15) 0·003 Percent intervention modules completed (% due) VHW report 2349/2748 (86%) 3392/3651 (93%) 4243/4713 (90%) 5214/5657 (92%) 9457/10370 (91%) 5741/6399 (90%) 0·42 8606/9308 (93%) 6592/7461 (88%) 0·025 Participant behaviours at 12 month visit Number of mothers with 12 month and 18 month outcomes Trial logs 113 119 131 158 289 232 .. 277 244 .. Number of children with 12 month and 18 month outcomes Trial logs 113 122 135 162 297 235 .. 284 248 ..

7/10370 (91%) 5741/6399 (90%) 0·42 8606/9308 (93%) 6592/7461 (88%) 0·025 Participant behaviours at 12 month visit Number of mothers with 12 month and 18 month outcomes Trial logs 113 119 131 158 289 232 .. 277 244 .. Number of children with 12 month and 18 month outcomes Trial logs 113 122 135 162 297 235 .. 284 248 .. WASH behaviours Household members who practice open defecation Maternal report 172/320 (54%) 183/442 (41%) 3/546 (1%) 0/699 3/1245 (<1%) 355/762 (47%) <0·001 NA NA .. Any latrine at household Observation 26/110 (24%) 45/120 (38%) 131/131 (100%) 155/155 (100%) 286/286 (100%) 71/230 (31%) <0·001 NA NA .. Improved latrine at household Observation 24/110 (22%) 33/120 (28%) 131/131 (100%) 154/154 (100%) 285/285 (100%) 57/230 (25%) <0·001 NA NA .. Improved latrine at household with well-trodden path, not used for storage, and not shared with other households Observation and maternal report 19/110 (17%) 23/120 (19%) 113/131 (86%) 134/154 (87%) 247/285 (87%) 42/230 (18%) <0·001 NA NA .. Handwashing station at household Observation 2/103 (2%) 7/117 (6%) 133/133 (100%) 157/158 (100%) 290/291 (100%) 9/220 (4%) <0·001 NA NA .. Handwashing station with water and rubbing agent at household Observation 1/102 (1%) 2/114 (2%) 101/122 (83%) 116/139 (84%) 217/261 (83%) 3/216 (1%) <0·001 NA NA .. Ever treats drinking water to make it safer Maternal report 9/109 (8%) 20/120 (17%) 107/130 (82%) 138/157 (85%) 245/287 (85%) 29/229 (13%) <0·001 NA NA .. Disposes rinse water from cleaning infant nappies with faeces in a latrine Maternal report 24/109 (22%) 37/115 (32%) 107/132 (81%) 114/145 (79%) 221/277 (80%) 61/224 (27%) <0·001 NA NA .. Play space is visibly clean Observation NA NA 117/127 (92%) 134/150 (89%) 251/277 (91%) NA .. NA NA .. Child ever observed to eat soil Maternal report 80/109 (73%) 75/121 (62%) 39/132 (30%) 26/154 (17%) 65/286 (23%) 155/230 (67%) <0·001 NA NA .. Child ever observed to eat chicken faeces Maternal report 21/109 (19%) 18/121 (15%) 3/132 (2%) 6/153 (4%) 9/285 (3%) 39/230 (17%) <0·001 NA NA .. IYCF behaviours Child is still breastfeeding Maternal report 100/112 (89%) 107/121 (88%) 119/134 (89%) 144/157 (92%) NA NA .. 251/278 (90%) 219/246 (89%) 0·66 Mother reports correct ways to feed child during and after illness Maternal report 83/112 (74%) 83/120 (69%) 92/132 (70%) 109/155 (70%) NA NA .. 192/275 (70%) 175/244 (72%) 0·65 Infant diet met minimum dietary diversity in past 24 h§ Maternal report 60/104 (58%) 73/119 (61%) 59/122 (48%) 106/146 (73%) NA NA ..

9%) 0·66 Mother reports correct ways to feed child during and after illness Maternal report 83/112 (74%) 83/120 (69%) 92/132 (70%) 109/155 (70%) NA NA .. 192/275 (70%) 175/244 (72%) 0·65 Infant diet met minimum dietary diversity in past 24 h§ Maternal report 60/104 (58%) 73/119 (61%) 59/122 (48%) 106/146 (73%) NA NA .. 179/265 (68%) 119/226 (53%) 0·002 Infant consumed iron rich food in the past 24 h§ Maternal report 64/110 (58%) 117/120 (98%) 61/131 (47%) 146/155 (94%) NA NA .. 263/275 (96%) 125/241 (52%) <0·001 Infant consumed animal source food in the past 24 h§ Maternal report 78/110 (71%) 87/120 (73%) 78/130 (60%) 114/152 (75%) NA NA .. 201/272 (74%) 156/240 (65%) 0·048 Infant consumed vitamin A rich food in the past 24 h§ Maternal report 72/111 (65%) 92/120 (77%) 95/134 (71%) 128/155 (83%) NA NA .. 220/275 (80%) 167/245 (68%) 0·002 SQ-LNS consumed in previous 24 h Maternal report NA 110/117 (94%) NA 128/155 (87%) NA NA .. 238/265 (90%) NA .. NA=not applicable. Data are n (%), unless otherwise indicated. SOC=standard of care. IYCF=infant and young child feeding. WASH=water, sanitation, and hygiene. IYCF plus WASH=both IYCF plus WASH implemented together. SQ-LNS=small-quantity lipid-based nutrient supplement. VHW=village health worker. * Combined WASH collapses the two WASH-containing groups (WASH and IYCF plus WASH); non-WASH collapses the two groups not including WASH (SOC and IYCF). † Combined IYCF collapses the two IYCF-containing groups (IYCF and IYCF plus WASH); non-IYCF collapses the two groups not including IYCF (SOC and WASH).

179/265 (68%) 119/226 (53%) 0·002 Infant consumed iron rich food in the past 24 h§ Maternal report 64/110 (58%) 117/120 (98%) 61/131 (47%) 146/155 (94%) NA NA .. 263/275 (96%) 125/241 (52%) <0·001 Infant consumed animal source food in the past 24 h§ Maternal report 78/110 (71%) 87/120 (73%) 78/130 (60%) 114/152 (75%) NA NA .. 201/272 (74%) 156/240 (65%) 0·048 Infant consumed vitamin A rich food in the past 24 h§ Maternal report 72/111 (65%) 92/120 (77%) 95/134 (71%) 128/155 (83%) NA NA .. 220/275 (80%) 167/245 (68%) 0·002 SQ-LNS consumed in previous 24 h Maternal report NA 110/117 (94%) NA 128/155 (87%) NA NA .. 238/265 (90%) NA .. NA=not applicable. Data are n (%), unless otherwise indicated. SOC=standard of care. IYCF=infant and young child feeding. WASH=water, sanitation, and hygiene. IYCF plus WASH=both IYCF plus WASH implemented together. SQ-LNS=small-quantity lipid-based nutrient supplement. VHW=village health worker. * Combined WASH collapses the two WASH-containing groups (WASH and IYCF plus WASH); non-WASH collapses the two groups not including WASH (SOC and IYCF). † Combined IYCF collapses the two IYCF-containing groups (IYCF and IYCF plus WASH); non-IYCF collapses the two groups not including IYCF (SOC and WASH). ‡ There were a maximum of 20 liquid soap deliveries, 15 WaterGuard deliveries, 13 SQ-LNS deliveries, 15 intervention modules.

* Combined WASH collapses the two WASH-containing groups (WASH and IYCF plus WASH); non-WASH collapses the two groups not including WASH (SOC and IYCF). † Combined IYCF collapses the two IYCF-containing groups (IYCF and IYCF plus WASH); non-IYCF collapses the two groups not including IYCF (SOC and WASH). ‡ There were a maximum of 20 liquid soap deliveries, 15 WaterGuard deliveries, 13 SQ-LNS deliveries, 15 intervention modules. § Calculations exclude SQ-LNS consumption. p values adjusted for clustering effect. Depending on the variable type, xtgee, multinomial, ordinal regression models with robust variance estimation, and Somers' D for medians, were used for comparing groups while handling within-cluster correlation.

‡ There were a maximum of 20 liquid soap deliveries, 15 WaterGuard deliveries, 13 SQ-LNS deliveries, 15 intervention modules. § Calculations exclude SQ-LNS consumption. p values adjusted for clustering effect. Depending on the variable type, xtgee, multinomial, ordinal regression models with robust variance estimation, and Somers' D for medians, were used for comparing groups while handling within-cluster correlation. Reported open defecation among household members was virtually eliminated in the WASH compared with the non-WASH groups and the vast majority had a latrine with a well-trodden path that was not being used for storage and a hand-washing station with soap or rubbing agent and water (table 2). At 12 months, 85% of women in the WASH groups reported they usually treat their drinking water. However, too few samples of water were tested for free chlorine to objectively validate water chlorination; we suspect uptake was modest: of 160 12-month water samples from WASH households that were tested, only 92 (58%) had >0·1ppm free chlorine. Fewer mothers in WASH households compared with non-WASH households reported ever seeing their child ingest soil or chicken faeces (table 2). Breastfeeding rates at 12 months were very high and did not differ across groups (table 2). The vast majority of children in the IYCF groups had consumed SQ-LNS in the previous 24 h. Compared with infants in the non-IYCF groups, a higher proportion in the IYCF groups met minimum dietary diversity, and had consumed foods that were animal source, iron rich, and vitamin-A rich the day before (table 2).

table 2). The vast majority of children in the IYCF groups had consumed SQ-LNS in the previous 24 h. Compared with infants in the non-IYCF groups, a higher proportion in the IYCF groups met minimum dietary diversity, and had consumed foods that were animal source, iron rich, and vitamin-A rich the day before (table 2). At the primary endpoint, median age of infants was 18·0 (IQR 17·8–18·6) months and did not differ significantly across treatment groups. The IYCF intervention significantly increased length for age Z score and haemoglobin concentration at 18 months (table 3). Length for age Z score was 0·26 higher (95% CI 0·09–0·43) and haemoglobin 2·9 g/L higher (95% CI 0·90–4·90) among children who received IYCF compared with those who did not. For both primary outcomes, the whole population was shifted upwards; there was no evidence of a greater effect of the IYCF intervention on the lower tails of length for age Z score and haemoglobin distributions (appendix). The IYCF intervention reduced the number of stunted children from 165 (50%) of 329 in the non-IYCF groups to 136 (40%) of 336 in the IYCF groups (absolute reduction 10%, 95% CI 2–17; RR 0·82, 95% CI 0·69–0·98). These findings were very similar in fully adjusted analyses (table 4). 45 (14%) of 319 children in the non-IYCF groups were anaemic at 18 months, compared with 24 (7%) of 329 in the IYCF groups (absolute difference 7%, 95% CI 2–12; RR 0·52, 95% CI 0·34–0·79). Results were attenuated in fully adjusted analyses, but comparisons between groups might have been limited by small numbers of events (table 4).Table 3 Effect of WASH and IYCF interventions on primary and secondary continuous outcomes at 18 months of age among children born to HIV-positive mothers

2, 95% CI 0·34–0·79). Results were attenuated in fully adjusted analyses, but comparisons between groups might have been limited by small numbers of events (table 4).Table 3 Effect of WASH and IYCF interventions on primary and secondary continuous outcomes at 18 months of age among children born to HIV-positive mothers Effects by group Main effects combining groups N Mean (SD) Treatment group N Mean (SD) Unadjusted difference (95% CI) p value Adjusted difference (95% CI)* p value Primary outcomes LAZ Standard of care 145 −2·00 (1·20) No IYCF 329 −1·99 (1·13) Ref .. Ref .. IYCF 146 −1·73 (1·10) IYCF 336 −1·73 (1·12) 0·26 (0·09 to 0.43) 0·003 0·23 (0·10 to 0·37) 0·001 WASH 184 −1·97 (1·08) No WASH 291 −1·87 (1·16) Ref .. Ref .. IYCF plus WASH 190 −1·73 (1·14) WASH 374 −1·85 (1·12) 0·01 (−0·16 to 0·18) 0·90 0·07 (−0·08 to 0·22) 0·37 Haemoglobin (g/L) Standard of care 141 116·8 (10·9) No IYCF 319 116·6 (12·9) Ref .. Ref .. IYCF 146 118·5 (11·2) IYCF 329 119·5 (11·7) 2·9 (0·90 to 4·90) 0·005 2·70 (0·60 to 4·80) 0·013 WASH 178 116·5 (14·3) No WASH 287 117·6 (11·1) Ref .. Ref .. IYCF plus WASH 183 120·3 (12·1) WASH 361 118·4 (13·3) 0·70 (−1·20 to 2·70) 0·47 1·10 (−0·90 to 3·20) 0·27 Secondary continuous outcomes WAZ Standard of care 146 −0·99 (1·10) No IYCF 329 −0·96 (1·14) Ref .. Ref .. IYCF 147 −0·96 (1·00) IYCF 336 −0·94 (1·06) 0·01 (−0·16 to 0·19) 0·87 0·02 (−0·17 to 0·15) 0·86 WASH 183 −0·93 (1·17) No WASH 293 −0·98 (1·05) Ref .. Ref .. IYCF plus WASH 189 −0·91 (1·10) WASH 372 −0·92 (1·13) 0·06 (−0·11 to 0·24) 0·47 0.05 (−0·12 to 0·21) 0·57 WHZ Standard of care 146 −0·05 (1·07) No IYCF 327 −0·05 (1·10) Ref .. Ref .. IYCF 147 −0·16 (1·08) IYCF 336 −0·13 (1·09) −0·09 (−0·27 to 0·09) 0·35 −0·08 (−0·27 to 0·10) 0·37 WASH 181 −0·04 (1·12) No WASH 293 −0·10 (1·08) Ref .. Ref .. IYCF plus WASH 189 −0·11 (1·10) WASH 370 −0·08 (1·11) 0·03 (−0·15 to 0·21) 0·71 −0·02 (−0·21 to 0·17) 0·83 Mid-upper arm circumference for age Z score Standard of care 146 −0·20 (0·94) No IYCF 328 −0·19 (0·88) Ref .. Ref .. IYCF 147 −0·20 (0·90) IYCF 337 −0·15 (0·93) 0·03 (−0·10 to 0·17) 0·63 −0·02 (−0·16 to 0·12) 0·76 WASH 182 −0·18 (0·83) No WASH 293 −0·20 (0·92) Ref .. Ref .. IYCF and WASH 190 −0·12 (0·94) WASH 372 −0·15 (0·89) 0·05 (−0·09 to 0·19) 0·50 0·06 (−0·08 to 0·21) 0·39 Head circumference Z score Standard of care 146 −0·55 (1·08) No IYCF 328 −0·54 (1·12) Ref .. Ref ..

) 0·03 (−0·10 to 0·17) 0·63 −0·02 (−0·16 to 0·12) 0·76 WASH 182 −0·18 (0·83) No WASH 293 −0·20 (0·92) Ref .. Ref .. IYCF and WASH 190 −0·12 (0·94) WASH 372 −0·15 (0·89) 0·05 (−0·09 to 0·19) 0·50 0·06 (−0·08 to 0·21) 0·39 Head circumference Z score Standard of care 146 −0·55 (1·08) No IYCF 328 −0·54 (1·12) Ref .. Ref .. IYCF 147 −0·51 (1·09) IYCF 336 −0·44 (1·12) 0·10 (−0·07 to 0·27) 0·24 0·08 (−0·08 to 0·25) 0·31 WASH 182 −0·53 (1·15) No WASH 293 −0·53 (1·09) Ref .. Ref .. IYCF and WASH 189 −0·38 (1·14) WASH 371 −0·46 (1·14) 0·07 (−0·10 to 0·24) 0·42 0·14 (−0·03 to 0·31) 0·098 IYCF=infant and young child feeding; WASH=water, sanitation and hygiene; IYCF plus WASH=both IYCF plus WASH implemented together. LAZ=length for age Z score. WAZ=weight for age Z score. WHZ=weight for height Z score. * Covariates included in adjusted analyses for LAZ and secondary growth outcomes were maternal height, maternal mid-upper arm circumference, marital status, maternal co-trimoxazole in pregnancy, low birthweight, infant sex, fieldworker, wealth quintile, household keeps livestock inside house, recruitment calendar period; covariates included in adjusted analyses for haemoglobin were maternal age, maternal haemoglobin, maternal employment, maternal ART in pregnancy, maternal co-trimoxazole in pregnancy, infant sex, fieldworker. Table 4 Effect of WASH and IYCF interventions on secondary dichotomous outcomes at 18 months of age among children born to HIV-positive mothers

* Covariates included in adjusted analyses for LAZ and secondary growth outcomes were maternal height, maternal mid-upper arm circumference, marital status, maternal co-trimoxazole in pregnancy, low birthweight, infant sex, fieldworker, wealth quintile, household keeps livestock inside house, recruitment calendar period; covariates included in adjusted analyses for haemoglobin were maternal age, maternal haemoglobin, maternal employment, maternal ART in pregnancy, maternal co-trimoxazole in pregnancy, infant sex, fieldworker. Table 4 Effect of WASH and IYCF interventions on secondary dichotomous outcomes at 18 months of age among children born to HIV-positive mothers Effects by group Main effects combining groups N n (%) Treatment group N n (%) Unadjusted relative risk (95% CI) p value Adjusted relative risk (95% CI) p value Stunting (LAZ <–2·0) SOC 145 75 (52%) No IYCF 329 165 (50%) Ref .. Ref .. IYCF 146 59 (40%) IYCF 336 136 (40%) 0·81 (0·68 to 0·97) 0·020 0·83 (0·71 to 0·99) 0·040 WASH 184 90 (49%) No WASH 291 134 (46%) Ref .. Ref .. IYCF plus WASH 190 77 (41%) WASH 374 167 (45%) 0·97 (0·81 to 1·15) 0·70 0·95 (0·80 to 1·12) 0·52 Severe stunting (LAZ <–3·0) Standard of care 145 22 (15%) No IYCF 329 51 (16%) Ref .. Ref .. IYCF 146 19 (13%) IYCF 336 42 (13%) 0·82 (0·55 to 1·23) 0·34 Insufficient sample NA WASH 184 29 (16%) No WASH 291 41 (14%) Ref .. Ref .. IYCF plus WASH 190 23 (12%) WASH 374 52 (14%) 0·97 (0·65 to 1·44) 0·87 Insufficient sample NA Anaemia (Haemoglobin <105 g/L) Standard of care 141 17 (12%) No IYCF 319 45 (14%) Ref .. Ref .. IYCF 146 7 (5%) IYCF 329 24 (7%) 0·52 (0·34 to 0·79) 0·002 0·95 (0·90 to 0·99) 0·033 WASH 178 28 (16%) No WASH 287 24 (8%) Ref .. Ref .. IYCF plus WASH 183 17 (9%) WASH 361 45 (13%) 1·42 (0·89 to 2·27) 0·14 1·01 (0·96 to 1·07) 0·57 Underweight (WAZ <–2·0) Standard of care 146 24 (16%) No IYCF 329 56 (17%) Ref .. Ref .. IYCF 147 27 (18%) IYCF 336 61 (18%) 1·07 (0·77 to 1·48) 0·70 Insufficient sample NA WASH 183 32 (18%) No WASH 293 51 (17%) Ref .. Ref .. IYCF plus WASH 189 34 (18%) WASH 372 66 (18%) 1·03 (0·74 to 1·42) 0·88 Insufficient sample NA Wasted (WHZ<–2·0) Standard of care 146 6 (4%) No IYCF 327 13 (4%) Ref .. Ref .. IYCF 147 6 (4%) IYCF 336 18 (5%) 1·30 (0·61 to 2·76) 0·49 Insufficient sample NA WASH 181 7 (4%) No WASH 293 12 (4%) Ref .. Ref .. IYCF plus WASH 189 12 (6%) WASH 370 19 (5%) 1·20 (0·56 to 2·57) 0·64 Insufficient sample NA IYCF=infant and young child feeding; WASH=water, sanitation and hygiene; IYCF plus WASH=both IYCF plus WASH implemented together. LAZ=length for age Z score. WAZ=weight for age Z score. WHZ=weight for height Z score.

f .. Ref .. IYCF plus WASH 189 12 (6%) WASH 370 19 (5%) 1·20 (0·56 to 2·57) 0·64 Insufficient sample NA IYCF=infant and young child feeding; WASH=water, sanitation and hygiene; IYCF plus WASH=both IYCF plus WASH implemented together. LAZ=length for age Z score. WAZ=weight for age Z score. WHZ=weight for height Z score. *Covariates included in adjusted analyses for secondary growth outcomes were maternal height, maternal mid-upper arm circumference, marital status, maternal co-trimoxazole in pregnancy, low birthweight, infant sex, fieldworker, wealth quintile, household keeps livestock inside house, recruitment calendar period; covariates included in adjusted analyses for secondary anaemia outcomes were maternal age, maternal haemoglobin, maternal employment, maternal ART in pregnancy, maternal co-trimoxazole in pregnancy, infant sex, fieldworker. Insufficient cases of severe anaemia to estimate relative risks.

e, recruitment calendar period; covariates included in adjusted analyses for secondary anaemia outcomes were maternal age, maternal haemoglobin, maternal employment, maternal ART in pregnancy, maternal co-trimoxazole in pregnancy, infant sex, fieldworker. Insufficient cases of severe anaemia to estimate relative risks. The WASH intervention had no effect on length for age Z score or haemoglobin concentrations. No difference was seen in mean length for age Z score (0·01, 95% CI −0·16 to 0·18) and mean haemoglobin concentration (0·7 g/L, 95% CI −1·20 to 2·70) at 18 months between children who received WASH and those who did not; effects were similar in adjusted analyses (table 3). There was no significant effect of the WASH intervention on stunting or anaemia (table 4). There were 134 (46%) of 291 stunted children in the non-WASH groups compared with 167 (45%) of 374 in the WASH groups (absolute difference 1%, 95% CI −9 to 6; RR 0·97, 95% CI 0·81–1·15), and there were 24 (8%) of 287 anaemic children in the non-WASH groups compared with 45 (12%) of 361 in the WASH groups (absolute difference 4%, 95% CI −1 to 9; RR 1·42, 95% CI 0·89–2·27). Neither intervention had an effect on other measures of growth (ie, weight for age, weight for height, mid-upper arm circumference, head circumference, or underweight or wasting; Table 3, Table 4); in a post-hoc analysis there was no effect of either intervention on the composite outcome of stunted and wasted: 7 (2%) of 335 children in the IYCF groups were stunted and wasted compared to 8 (2%) of 326 in the non-IYCF groups (absolute difference 0%; 95% CI −3 to 2); and 9 (2%) of 370 children in the WASH groups were stunted and wasted compared to 6 (2%) of 291 in the non-WASH groups (absolute difference 0%; 95% CI −2 to 3).

d: 7 (2%) of 335 children in the IYCF groups were stunted and wasted compared to 8 (2%) of 326 in the non-IYCF groups (absolute difference 0%; 95% CI −3 to 2); and 9 (2%) of 370 children in the WASH groups were stunted and wasted compared to 6 (2%) of 291 in the non-WASH groups (absolute difference 0%; 95% CI −2 to 3). There were no significant differences in 7-day prevalence of diarrhoea between groups at 12 months (appendix) or 18 months (table 5). The numbers of children with dysentery and acute respiratory infection were too few to compare. Cumulative mortality through 18 months was not significantly different between groups. In the standard of care group, 13 (8%) of 165 children died, compared with 6 (4%) of 156 in the IYCF group, 14 (7%) of 205 in the WASH group, and 15 (7%) of 207 in the IYCF plus WASH group. Adjusted analyses were not done owing to the small absolute numbers of cases.Table 5 Effect of IYCF and WASH on diarrhoea, dysentery, acute respiratory infection, and mortality at 18 months

, compared with 6 (4%) of 156 in the IYCF group, 14 (7%) of 205 in the WASH group, and 15 (7%) of 207 in the IYCF plus WASH group. Adjusted analyses were not done owing to the small absolute numbers of cases.Table 5 Effect of IYCF and WASH on diarrhoea, dysentery, acute respiratory infection, and mortality at 18 months Effects by group Main effects combining groups N n (%) Treatment group N n (%) Unadjusted risk ratio*(95% CI) p value Diarrhoea Standard of care 145 9 (6%) No IYCF 328 20 (6%) Reference .. IYCF 147 13 (9%) IYCF 336 23 (7%) 1·12 (0·62–2·03) 0·72 WASH 183 11 (6%) No WASH 292 22 (8%) Reference .. IYCF plus WASH 189 10 (5%) WASH 372 21 (6%) 0·73 (0·40–1·33) 0·31 Dysentery Standard of care 145 2 (1%) No IYCF 326 4 (1%) Reference .. IYCF 147 0 IYCF 336 0 Insufficient sample NA WASH 181 2 (1%) No WASH 292 2 (1%) Reference .. IYCF plus WASH 189 0 WASH 370 2 (<1%) Insufficient sample NA Acute respiratory infection Standard of care 145 0 No IYCF 327 2 (1%) Reference .. IYCF 146 1 (1%) IYCF 335 4 (1%) Insufficient sample NA WASH 182 2 (1%) No WASH 291 1 (<1%) Reference .. IYCF plus WASH 189 3 (2%) WASH 371 5 (1%) Insufficient sample NA Death Standard of care 165 13 (8%) No IYCF 370 27 (7%) Reference .. IYCF 156 6 (4%) IYCF 363 21 (6%) 0·81 (0·44–1·49) 0·49 WASH 205 14 (7%) No WASH 321 19 (6%) Reference .. IYCF plus WASH 207 15 (7%) WASH 412 29 (7%) 1·26 (0·68–2·34) 0·47 IYCF=infant and young child feeding. WASH=water, sanitation, and hygiene. NA=not applicable. All outcomes apart from death were based on maternal 7-day recall. Diarrhoea was defined as passage of three or more loose or watery stools in a 24 h period. Dysentery was defined as passage of stool with blood or mucus. Acute respiratory infection was defined as fast or difficult breathing (ie, rapid breathing or chest retractions, or both).

death were based on maternal 7-day recall. Diarrhoea was defined as passage of three or more loose or watery stools in a 24 h period. Dysentery was defined as passage of stool with blood or mucus. Acute respiratory infection was defined as fast or difficult breathing (ie, rapid breathing or chest retractions, or both). * Adjusted analyses were not done owing to the small absolute numbers of cases.

death were based on maternal 7-day recall. Diarrhoea was defined as passage of three or more loose or watery stools in a 24 h period. Dysentery was defined as passage of stool with blood or mucus. Acute respiratory infection was defined as fast or difficult breathing (ie, rapid breathing or chest retractions, or both). * Adjusted analyses were not done owing to the small absolute numbers of cases. In the prespecified per-protocol analysis restricted to those with high-fidelity delivery of trial interventions, effects of IYCF and WASH on growth were similar to the intention-to-treat findings, but there was also a significant effect of IYCF on mean head circumference for age Z score (0·20 [95% CI 0·01–0·39] higher in IYCF compared with non-IYCF groups, p=0·037; appendix). The effects of IYCF on haemoglobin concentration and anaemia were more pronounced in the per-protocol analysis than in the intention-to-treat analysis: haemoglobin concentration was 4·3 g/L (95% CI 2·0–6·5) higher in IYCF compared with non-IYCF groups, and there were fewer anaemic children in the IYCF groups (14 [6%] of 253) than in the non-IYCF groups (39 [17%] of 236; absolute difference 11%, 95% CI 5–17; RR 0·35, 95% CI 0·22–0·56). The WASH intervention significantly reduced 7-day prevalence of diarrhoea at the 18-month visit in the per-protocol analysis: 17 (8%) of 209 children in the non-WASH groups had diarrhoea, compared with 10 (3%) of 290 in the WASH groups (absolute difference 5%, 95% CI 0–9); RR 0·42, 95% CI 0·21–0·85), but had no effect at 12 months or on other measures of morbidity. In a preplanned subgroup analysis, infant sex did not modify the effects of IYCF or WASH on either primary outcome (interactions all p>0·10).

ompared with 10 (3%) of 290 in the WASH groups (absolute difference 5%, 95% CI 0–9); RR 0·42, 95% CI 0·21–0·85), but had no effect at 12 months or on other measures of morbidity. In a preplanned subgroup analysis, infant sex did not modify the effects of IYCF or WASH on either primary outcome (interactions all p>0·10). When the effects of the interventions were limited only to children confirmed as HIV-uninfected (ie, removing children who were HIV-positive and HIV-unknown at 18 months), the overall findings were similar (appendix). Serious adverse events were similar across trial groups (table 6). There were no serious adverse events or adverse events related to the trial interventions.Table 6 Cumulative distribution of serious adverse events among HIV-positive women and HIV-exposed infants by randomised trial arm Standard of care IYCF WASH IYCF plus WASH Miscarriages (n/N) 5/174 (3%) 13/182 (7%) 11/213 (5%) 5/215 (2%) Stillbirths (n/N) 2/174 (1%) 13/186 (15%) 2/219 (1%) 4/219 (2%) Neonatal deaths (<1month; n/N) 6/174 (3%) 3/186 (2%) 9/219 (4%) 9/219 (4%) Infant deaths (n/N) 8/174 (5%) 4/186 (2%) 5/219 (2%) 7/219 (3%) Maternal hospitalisation (n/N) 8/174 (5%) 15/182 (8%) 7/213 (3%) 22/215 (10%) Infant hospitalisation (n/N) 2/174 (1%) 7/186 (4%) 5/219 (2%) 2/219 (1%)

86 (15%) 2/219 (1%) 4/219 (2%) Neonatal deaths (<1month; n/N) 6/174 (3%) 3/186 (2%) 9/219 (4%) 9/219 (4%) Infant deaths (n/N) 8/174 (5%) 4/186 (2%) 5/219 (2%) 7/219 (3%) Maternal hospitalisation (n/N) 8/174 (5%) 15/182 (8%) 7/213 (3%) 22/215 (10%) Infant hospitalisation (n/N) 2/174 (1%) 7/186 (4%) 5/219 (2%) 2/219 (1%) Discussion We evaluated the individual and combined effects of IYCF and improved WASH on child linear growth and haemoglobin concentrations among HIV-exposed infants, in whom stunting and anaemia are common. HIV-exposed infants have an excess risk of stunting and therefore might particularly benefit from public health interventions aimed at promoting healthy growth. We found that IYCF had a significant but modest effect on linear growth, reducing stunting by almost 20%, whereas there were no clear benefits from the household WASH intervention. IYCF also increased haemoglobin and reduced the prevalence of anaemia, but neither intervention consistently affected morbidity or all-cause mortality. Overall, these findings are similar to those reported previously among HIV-unexposed children,11 and demonstrate that in settings of high antenatal HIV prevalence, IYCF interventions would have substantial benefits at scale, whereas integrating household-level elementary WASH interventions typical of those commonly available to rural populations in low-income and middle-income countries (ie, pit latrines, hand-washing stations not connected to a water source, and point-of-use drinking water with monthly behaviour-change communication) is unlikely to confer additional benefits on child growth or anaemia.

typical of those commonly available to rural populations in low-income and middle-income countries (ie, pit latrines, hand-washing stations not connected to a water source, and point-of-use drinking water with monthly behaviour-change communication) is unlikely to confer additional benefits on child growth or anaemia. Linear growth benefits among HIV-exposed uninfected infants could be particularly important because of the high risk of stunting in this vulnerable population: in the non-IYCF groups, 50% of HIV-exposed uninfected infants were stunted by 18 months, despite high uptake of interventions to prevent mother-to-child transmission. Although the effect of IYCF on linear growth was modest (increased length for age Z score by 0·26), it exceeded the effect size observed in non-HIV-exposed children in SHINE11 (0·16 length for age Z score increase) and the average effect size (0·10 length for age Z score increase) reported in a recent meta-analysis of complementary feeding interventions among predominantly HIV-unexposed infants.17 There might be differences in household food security, dietary diversity, and caregiving practices between infants born to HIV-positive and HIV-negative mothers, such that HIV-exposed uninfected children showed particular benefits for linear growth from provision of additional calories, micronutrients, and education focused on diversification and calorie enrichment of infant diets. IYCF did not reduce the prevalence of underweight or wasting in HIV-exposed children. The effect of IYCF on haemoglobin was also modest (2·9 g/L gain at 18 months) but translated into a substantial reduction in the proportion of anaemic children at 18 months. HIV-exposed infants have a higher frequency of anaemia than do HIV-unexposed infants,18, 19 and might respond better to iron supplementation, although the causes of anaemia in HIV-exposed uninfected infants remain unclear. Notably, all the trial findings were similar after excluding HIV-infected children and children of unknown HIV status.

ave a higher frequency of anaemia than do HIV-unexposed infants,18, 19 and might respond better to iron supplementation, although the causes of anaemia in HIV-exposed uninfected infants remain unclear. Notably, all the trial findings were similar after excluding HIV-infected children and children of unknown HIV status. The WASH intervention had no effect on growth or anaemia and inconsistent effects on diarrhoea in HIV-exposed children. Our hypothesis was that WASH would reduce diarrhoea and prevent environmental enteric dysfunction, which would, in turn, reduce stunting by reducing malabsorption and chronic inflammation. We reasoned that HIV-exposed infants might be particularly vulnerable to environmental enteric dysfunction, owing to perturbed composition and function of the microbiota, which is vertically transmitted from an HIV-positive mother,20 and direct effects of HIV exposure during breastfeeding on the gut barrier and intestinal mucosal CD4 cells.21 However, we found no effect of the WASH intervention on linear growth, and inconsistent reductions in diarrhoea between the intention-to-treat and per-protocol populations. Uptake of the WASH intervention was high, as assessed by structured observations and self-report, although the intervention intensity might have been too low to modify household behaviours to the extent necessary to affect these health outcomes.11 We previously speculated that the elementary WASH interventions implemented in the SHINE trial might not have been effective enough to reduce highly contaminated environments. We believe that so-called transformative WASH is urgently required, which combines better tools, more intensive behaviour change, and strengthened governance for implementing these interventions.11 Ongoing laboratory studies will determine whether the WASH interventions affected any biomarkers of environmental enteric dysfunction.

hat so-called transformative WASH is urgently required, which combines better tools, more intensive behaviour change, and strengthened governance for implementing these interventions.11 Ongoing laboratory studies will determine whether the WASH interventions affected any biomarkers of environmental enteric dysfunction. The WASH Benefits trials in Kenya22 and Bangladesh23 have recently reported the effects of IYCF and WASH interventions on growth and anaemia. SHINE was purposefully aligned with WASH Benefits in design, but was the only trial carried out in a setting with high HIV prevalence (15% in SHINE, <6% in WASH Benefits Kenya,22 and <0·1% in WASH Benefits Bangladesh24); we therefore tested mothers and infants for HIV and stratified results by maternal HIV status. Findings on stunting and anaemia were similar across the three trials, despite the differences in context and populations: all three trials showed benefits of IYCF but none found an effect of the elementary household-level WASH interventions tested on stunting and anaemia. Findings to date show that the IYCF benefits might be greater in HIV-exposed infants than in HIV-unexposed infants, providing important evidence that an IYCF intervention is likely to be especially effective in populations with high antenatal HIV prevalence. However, elementary household-level WASH interventions—unaccompanied by investments in more intensive behaviour-change communication, efficacious technologies, and strengthened governance systems of financing, regulation, and management—are unlikely to reduce child stunting or anaemia.

th high antenatal HIV prevalence. However, elementary household-level WASH interventions—unaccompanied by investments in more intensive behaviour-change communication, efficacious technologies, and strengthened governance systems of financing, regulation, and management—are unlikely to reduce child stunting or anaemia. Our trial had several strengths and limitations. We delivered household-level public health interventions with high fidelity of implementation and created substantial contrast in hardware, commodities and behaviours between groups. The level of input was higher than would be provided in a typical nutrition or WASH programme delivered to rural areas of low-income countries. We were unable to blind interventions to participants or fieldworkers because of the nature of the household inputs; however, investigators were masked during analysis and the consistency of findings across both primary outcomes suggests this factor is unlikely to have caused substantial ascertainment bias. Despite our use of constrained randomisation, which balanced clusters on a range of variables, there were some baseline differences between groups; however, we did adjusted analyses to accommodate for this imbalance. Adjustment mostly led to very similar findings, except for the reduction in anaemia, which was significantly attenuated in the fully adjusted analysis. The finding for anaemia might be at least partly explained by a relatively small absolute number of anaemia cases (table 4). We were unable to determine HIV status for all children at 18 months because of caregiver refusal for blood draws, insufficient sample volume, or discordant results; however, in sensitivity analyses that removed children with unknown or positive HIV status, the effect of the interventions was unchanged. The trial was conducted in a community with high, but not universal, uptake of interventions to prevent mother-to-child transmission; however, we intervened in the trial to promote exclusive breastfeeding to high levels,25 which might have reduced HIV transmission26 and improved infant health outcomes regardless of randomised interventions. Finally, the sample size for the trial was based on detecting a difference in length for age Z score among HIV-unexposed infant groups; we did not calculate a specific sample size for HIV-exposed infants.

h might have reduced HIV transmission26 and improved infant health outcomes regardless of randomised interventions. Finally, the sample size for the trial was based on detecting a difference in length for age Z score among HIV-unexposed infant groups; we did not calculate a specific sample size for HIV-exposed infants. It is possible that the null effect of the WASH intervention was due to insufficient power to detect an effect; however, we think this is unlikely given that there was no evidence of a difference in length for age Z score between the WASH and non-WASH groups in the much larger sample of 3686 HIV-unexposed children in whom we have previously reported findings.11

WASH intervention was due to insufficient power to detect an effect; however, we think this is unlikely given that there was no evidence of a difference in length for age Z score between the WASH and non-WASH groups in the much larger sample of 3686 HIV-unexposed children in whom we have previously reported findings.11 In summary, a complementary feeding intervention that promoted behaviour change and provided a daily calorie and micronutrient supplement in the form of a lipid-based nutrient supplement improved linear growth and haemoglobin concentration in HIV-exposed children at 18 months old. A household-level WASH intervention providing elementary WASH tools and monthly behaviour-change communication had no evidence of an effect on growth or anaemia, and no consistent effect on diarrhoea. These findings are similar to those observed in the HIV-unexposed children enrolled in SHINE. These findings are important because stunting and anaemia are major causes of morbidity, mortality, and impaired neurodevelopment globally,8 and interventions that improve linear growth and haemoglobin might confer benefits across the life course. Although the absolute effect of the IYCF intervention on mean length for age Z score and haemoglobin was modest, it led to substantial reductions in stunting and anaemia by 18 months of age, which might have considerable public health benefit at scale. Overall, our findings highlight the need for greater coverage of IYCF interventions, particularly in areas of ongoing high antenatal HIV prevalence, where they might be particularly beneficial, and provide an urgent call for WASH interventions that are more efficacious.

have considerable public health benefit at scale. Overall, our findings highlight the need for greater coverage of IYCF interventions, particularly in areas of ongoing high antenatal HIV prevalence, where they might be particularly beneficial, and provide an urgent call for WASH interventions that are more efficacious. For the protocol and statistical analysis plan see https://osf.io/w93hy For interactive tools see https://osf.io/w93hy For the protocol and statistical analysis plan see https://osf.io/w93hy For the Clinical Epidemiology Database Resources see http://ClinEpiDB.org Supplementary Material Supplementary appendix

have considerable public health benefit at scale. Overall, our findings highlight the need for greater coverage of IYCF interventions, particularly in areas of ongoing high antenatal HIV prevalence, where they might be particularly beneficial, and provide an urgent call for WASH interventions that are more efficacious. For the protocol and statistical analysis plan see https://osf.io/w93hy For interactive tools see https://osf.io/w93hy For the protocol and statistical analysis plan see https://osf.io/w93hy For the Clinical Epidemiology Database Resources see http://ClinEpiDB.org Supplementary Material Supplementary appendix Acknowledgments We thank all the mothers, babies, and their families who participated in SHINE. We gratefully acknowledge the leadership and staff of the Ministry of Health and Child Care in Chirumanzu and Shurugwi districts and Midlands Province (especially environmental health, nursing, and nutrition) for their roles in operationalisation of the study procedures. We acknowledge the Ministry of Local Government officials in each district who supported and facilitated field operations. We are particularly indebted to Phillipa Rambanepasi and her team for proficiently managing all the finances and Virginia Sauramba for managing compliance issues. Finally, we are very thankful for our programme officers at the Bill & Melinda Gates Foundation and UK Aid who enthusiastically worked with us over a long period of time to make SHINE happen. The SHINE trial is funded by the Bill & Melinda Gates Foundation (OPP1021542 and OPP113707); UK Department for International Development; Wellcome Trust, UK (093768/Z/10/Z, 108065/Z/15/Z and 203905/Z/16/Z); Swiss Agency for Development and Cooperation; US National Institutes of Health (2R01HD060338-06); and UNICEF (PCA-2017-0002).

ial is funded by the Bill & Melinda Gates Foundation (OPP1021542 and OPP113707); UK Department for International Development; Wellcome Trust, UK (093768/Z/10/Z, 108065/Z/15/Z and 203905/Z/16/Z); Swiss Agency for Development and Cooperation; US National Institutes of Health (2R01HD060338-06); and UNICEF (PCA-2017-0002). Data sharing statement The full protocol and statistical analysis plan for the SHINE trial are available at https://osf.io/w93hy from March 27, 2018. Data collected for the SHINE trial will be made publicly available as individual participant data with an accompanying data dictionary. The data will be housed and made accessible to the global research community through Clinical Epidemiology Database Resources at the University of Pennsylvania. This platform is charged with ensuring that epidemiological studies are fully anonymised by removing all personal identifiers and obfuscating all dates per participant through application of a random number algorithm to comply with the ethical conduct of human subjects research. Researchers must agree to the policies and comply with the mechanism of ClinEpiDB to access data housed on this platform.

nymised by removing all personal identifiers and obfuscating all dates per participant through application of a random number algorithm to comply with the ethical conduct of human subjects research. Researchers must agree to the policies and comply with the mechanism of ClinEpiDB to access data housed on this platform. Contributors AJP directed all clinical and laboratory aspects of the trial and managed the data collection and laboratory teams. BC, CE, and LES contributed to data analysis and interpretation of SHINE. MNNM led the development of the interventions and managed their implementation. RN developed and managed all information technology, data management, and data analysis. RJS co-authored the original protocols and contributed to design and implementation of the trial and data analysis and interpretation. LHM was the senior statistician. NVT managed field operations. KM managed the laboratory and led all HIV testing and interpretation. FDM and BM supervised all data collection. CMC and AC contributed to the trial design and served as liaisons to the departments of nursing and nutrition, respectively, in the Ministry of Health and Child Care, Zimbabwe. and JHH and GTM were the co-principal investigators of the trial. All authors contributed to, reviewed, and approved this manuscript. Declaration of interests We declare no competing interests.

Introduction Adolescence, a period of crucial physical and neuro-maturational changes, is increasingly recognised as a life stage worthy of strategic health investments.1 The UN's Sustainable Development Agenda includes specific targets on reducing poverty and hunger, ensuring healthy lives and wellbeing at all ages, empowering women, and achieving equity in education; these targets will not be reached without building the capabilities of adolescents.2 South Asia has the largest number of stunted children globally. It is also home to the most adolescents of any global region, and one in five adolescents globally live in India. Among the many health issues that adolescents face, teenage pregnancy is arguably of the greatest consequence due to its effects on the wellbeing of both the mother and child. An estimated 16 million girls aged 15–19 years give birth annually, and 95% of these births occur in low-income and middle-income countries.3 Although marriage before the age of 18 years has been illegal in India since 1929—with the law updated as the Prohibition of Child Marriage Act in 2006—girls, especially those from poor rural areas, continue to be married early. Furthermore, owing to societal pressure to consummate the marriage and low sexual reproductive health knowledge, among other factors, 31% of married Indian women gave birth by the age of 18 years in 2016.4 Research in context Evidence before this study

South Asia has the largest number of stunted children globally. It is also home to the most adolescents of any global region, and one in five adolescents globally live in India. Among the many health issues that adolescents face, teenage pregnancy is arguably of the greatest consequence due to its effects on the wellbeing of both the mother and child. An estimated 16 million girls aged 15–19 years give birth annually, and 95% of these births occur in low-income and middle-income countries.3 Although marriage before the age of 18 years has been illegal in India since 1929—with the law updated as the Prohibition of Child Marriage Act in 2006—girls, especially those from poor rural areas, continue to be married early. Furthermore, owing to societal pressure to consummate the marriage and low sexual reproductive health knowledge, among other factors, 31% of married Indian women gave birth by the age of 18 years in 2016.4 Research in context Evidence before this study Childhood stunting is highly prevalent in low-income countries, and children born to adolescent mothers are more likely to be stunted compared with those born to adult mothers. We searched PubMed from database inception to Dec 19, 2018, with filters for English language and human studies and the following search terms: (adolescent OR adolescence OR teenage OR teen) AND (pregnant OR pregnancy) AND (infant OR child) AND (stunting OR stunted). We identified 27 studies that examined the relationship between adolescent pregnancy and childhood stunting. Different studies separately identified relationships related to women's nutrition, access to health services during antenatal, delivery, and early childhood periods, infant and young child feeding, living conditions, and women's education and bargaining power. No studies had empirically examined all seven groups of factors together; the most comprehensive empirical study was from Bangladesh, reporting on four of these seven factors.

antenatal, delivery, and early childhood periods, infant and young child feeding, living conditions, and women's education and bargaining power. No studies had empirically examined all seven groups of factors together; the most comprehensive empirical study was from Bangladesh, reporting on four of these seven factors. Added value of this study To our knowledge, this is the most comprehensive study to date of biological, social, and programmatic factors at multiple levels (individual, household, and health services) that aims to explain the detrimental effects of early childbearing age on child stunting. By use of nationally representative data on more than 60 000 mother-child pairs in India and examination of multiple pathways, we found that adolescent pregnancy is associated with child undernutrition through factors such as poorer maternal nutritional status, lower educational attainment, less access to health services during antenatal or postnatal care and early childhood, suboptimal complementary feeding practices, and poorer living conditions compared with adult pregnancy. Together, these factors accounted for an 11 percentage point higher prevalence of child stunting. Implications of all the available evidence

To our knowledge, this is the most comprehensive study to date of biological, social, and programmatic factors at multiple levels (individual, household, and health services) that aims to explain the detrimental effects of early childbearing age on child stunting. By use of nationally representative data on more than 60 000 mother-child pairs in India and examination of multiple pathways, we found that adolescent pregnancy is associated with child undernutrition through factors such as poorer maternal nutritional status, lower educational attainment, less access to health services during antenatal or postnatal care and early childhood, suboptimal complementary feeding practices, and poorer living conditions compared with adult pregnancy. Together, these factors accounted for an 11 percentage point higher prevalence of child stunting. Implications of all the available evidence The findings of this study, together with previous studies, show that adolescent pregnancy is linked to childhood stunting through a wide set of factors. As one of the ten countries with the largest burden of early childbearing in both relative prevalence and absolute number—and the country with the most stunted children—reducing adolescent pregnancy in India can likely help achieve several Sustainable Development Goals related to poverty, health, nutrition, general wellbeing, equity, and education. Policies and programmes to delay childbearing have the potential to help break the intergenerational cycle of poverty and undernutrition through broad effects on multiple determinants of childhood stunting.

al Sustainable Development Goals related to poverty, health, nutrition, general wellbeing, equity, and education. Policies and programmes to delay childbearing have the potential to help break the intergenerational cycle of poverty and undernutrition through broad effects on multiple determinants of childhood stunting. The adverse consequences of early childbearing on maternal and child health and wellbeing are far ranging.5, 6, 7, 8 Pregnancy and childbirth complications are the leading cause of death among 15–19-year-old girls globally.9 Adolescent pregnancy often results in school dropout, affecting young women's education and income.1, 10 Women's nutritional status is also affected.11, 12 The prevalence of thinness among Indian women is twice as high in those married before 18 years of age than those married after 24 years of age (33% vs 16%).13 Furthermore, women who become pregnant as adolescents might not have access to high quality health services during the crucial first 1000-day period,14, 15 which might have long-term consequences for their children.5, 7

igh in those married before 18 years of age than those married after 24 years of age (33% vs 16%).13 Furthermore, women who become pregnant as adolescents might not have access to high quality health services during the crucial first 1000-day period,14, 15 which might have long-term consequences for their children.5, 7 Despite evidence that pregnancy during adolescence negatively affects maternal and child outcomes, to our knowledge no studies have taken a holistic approach to understanding how early childbearing is linked to child undernutrition. Although studies have examined some potential pathways, few have empirically examined multiple biological, social, and programmatic factors. Understanding these relationships can help to build support for policies and programmes that improve adolescent, maternal, and child health and wellbeing. We aimed to determine to what extent adolescent pregnancy is associated with poor child nutritional outcomes and through which pathways adolescent pregnancy is linked to child undernutrition.

onships can help to build support for policies and programmes that improve adolescent, maternal, and child health and wellbeing. We aimed to determine to what extent adolescent pregnancy is associated with poor child nutritional outcomes and through which pathways adolescent pregnancy is linked to child undernutrition. Methods Data sources This Article uses data from the fourth round of India's National Family Health Survey (NFHS-4)16 which is representative at both the state and district levels, gathering data from 601 509 households. To avoid biases associated with parity and birth spacing, we only included primiparous women aged 15–49 years who had given birth in the previous 5 years.15 Women were classified on the basis of their age at the time of their first livebirth: 10–19 years (adolescents), 20–24 years (young adults), and 25 years or older (adults). Details of sample selection are provided in the appendix. Outcomes The primary outcomes in this study were child anthropometric measures. Children's weight and length or height measurements were used to derive Z scores by comparing each child's anthropometric measurements with the WHO age-appropriate and sex-appropriate child growth standards.17 Three indicators were calculated: length or height-for-age Z score (HAZ), weight-for-age Z score (WAZ), and weight-for-length or height Z score (WHZ). Stunting was defined as a HAZ of less than −2, underweight as a WAZ of less than −2, and wasting as a WHZ of less than −2.17

nd sex-appropriate child growth standards.17 Three indicators were calculated: length or height-for-age Z score (HAZ), weight-for-age Z score (WAZ), and weight-for-length or height Z score (WHZ). Stunting was defined as a HAZ of less than −2, underweight as a WAZ of less than −2, and wasting as a WHZ of less than −2.17 Framework of determinants We were not able to find any comprehensive frameworks to explain potential associations between adolescent pregnancy and child undernutrition. Therefore, we developed an evidence-based conceptual framework on the basis of a literature review to guide our path analyses (figure 1). We did a systematic review of the global published literature linking adolescent pregnancy to child stunting. The search details and relevant studies are presented in the appendix. We identified 27 studies published between 1990 and 2018; from these studies we identified five broad groups of factors linking adolescent pregnancy to child undernutrition: maternal nutritional status, education and bargaining power, access to health services, child feeding practices, and living conditions. We used available variables in the NFHS-4 to empirically test the role of these factors, recognising that a few variables reflect the status of the mother at the time of the survey rather than the status at the time of her pregnancy. Details of the indicators used are outlined in the panel. Although we show directionality using arrows in figure 1 for illustrative purposes of early pregnancy possibly leading to child undernutrition, we acknowledge the interplay of factors presented in the framework, such as the bidirectional nature of the links between adolescent pregnancy and maternal education, bargaining power, and living conditions.Figure 1 Conceptual framework for linking adolescent pregnancy and early childhood undernutrition

undernutrition, we acknowledge the interplay of factors presented in the framework, such as the bidirectional nature of the links between adolescent pregnancy and maternal education, bargaining power, and living conditions.Figure 1 Conceptual framework for linking adolescent pregnancy and early childhood undernutrition Panel Indicators included in analyses linking adolescent pregnancy to child undernutrition Maternal nutritional status Maternal weight and height were used to calculate body-mass index (BMI), with low BMI defined as <18·5 kg/m2. Haemoglobin concentrations were measured from capillary blood samples, using a portable HemoCue Hb 201 + analyser. Haemoglobin concentration was adjusted for cigarette smoking and for altitude in areas above 1000 m, and anaemia was defined as a concentration of less than 120 g/L for non-pregnant women.18 Access to nutrition and health services Antenatal care services To assess antenatal health services, we included early antenatal care visit (received antenatal care during the first trimester of pregnancy), at least four antenatal care visits, received iron and folic acid supplements, and received deworming during pregnancy. Delivery and postnatal health services To assess delivery and postnatal health services, we included institutional delivery, skilled birth attendant present during delivery, and postnatal care for mothers during the first 2 days after birth. Early childhood health services

To assess antenatal health services, we included early antenatal care visit (received antenatal care during the first trimester of pregnancy), at least four antenatal care visits, received iron and folic acid supplements, and received deworming during pregnancy. Delivery and postnatal health services To assess delivery and postnatal health services, we included institutional delivery, skilled birth attendant present during delivery, and postnatal care for mothers during the first 2 days after birth. Early childhood health services Indicators related to early childhood included growth monitoring, food supplementation (among children ≥6 months of age), full immunisation (among children ≥12 months of age), paediatric iron and folic acid and vitamin A supplementation, and deworming (among children ≥6 months of age). Infant and young child feeding Infant and young child feeding practices were assessed using the standard WHO indicators,19 including early initiation of breastfeeding (the proportion of infants breastfed within 1 h of birth), exclusive breastfeeding (the proportion of infants 0–5·9 months of age who were fed only breast milk), adequate diet (defined as children who consumed at least four of seven food groups in the previous 24 h and age-appropriate meal frequency), and consumption of iron-rich food.19 Living conditions

regnant during school.21, 22 Furthermore, maternal education is known to be a strong determinant of many child outcomes, including linear growth.23 Women's bargaining power might still be an important determinant of child nutrition; however the effects might have been outweighed by education in our multivariable model. Use of health services during antenatal period was an important link between adolescent pregnancy and child nutrition. Our results match findings from another study15 in west Africa showing that adolescent (aged 10–19 years) mothers seek care later, make fewer visits during pregnancy, and receive fewer components of care than older first-time mothers. Our work extends these findings by examining implications during delivery and postnatal periods as well as at the child level. Whereas most antenatal health service factors were important in our path models, we found that receipt of health services during postnatal and early childhood periods had less of a role. Although adolescent pregnancy was associated with poorer access to services during the postpartum and early childhood periods, these indicators were not significantly associated with child HAZ in our multivariable regression model after controlling for maternal education and living conditions.

Infant and young child feeding practices were assessed using the standard WHO indicators,19 including early initiation of breastfeeding (the proportion of infants breastfed within 1 h of birth), exclusive breastfeeding (the proportion of infants 0–5·9 months of age who were fed only breast milk), adequate diet (defined as children who consumed at least four of seven food groups in the previous 24 h and age-appropriate meal frequency), and consumption of iron-rich food.19 Living conditions A household socioeconomic status index was constructed using principal component analysis, extracting from multiple variables including house and land ownership, housing structure, and ownership of assets and livestock.20 Sanitation was captured by whether the household had an improved sanitation facility. Education and bargaining power Women's status and bargaining power were measured through their education (completed years of education), work for pay in the last 12 months, ownership of money and large assets (land or house), having a say in household decisions (a composite score of decisions on health care, large household purchases, ability to spend the husband's earnings, and whether permission is needed to visit family or relatives), and mobility (a composite score of women's ability to travel alone to the market, to the health facility, and out of the village).

decisions (a composite score of decisions on health care, large household purchases, ability to spend the husband's earnings, and whether permission is needed to visit family or relatives), and mobility (a composite score of women's ability to travel alone to the market, to the health facility, and out of the village). Statistical analysis We calculated women's age at first birth as the difference between the birth date of the first-born child and the birth date of the woman. We estimated the proportion of women in the sample who were adolescents, young adults, and adults at the time of their first livebirth. We used survey analysis procedures in Stata version 15 to account for the cluster sampling design and sampling weights in the NFHS-4 in all analyses described here, including these estimates. To determine the extent to which adolescent pregnancy is associated with poor child nutritional outcomes, we first visualised how child anthropometric outcomes differ among children born to adolescents, young adults, or adult women. We plotted HAZ, WAZ, and WHZ by child age from 0 to 60 months using smoothed polynomial plots. We then compared these outcomes among the groups using multivariable regression models, adjusting for child age, sex, mother's religion, and scheduled caste or tribe (designated groups of historically disadvantaged people in India).

ted HAZ, WAZ, and WHZ by child age from 0 to 60 months using smoothed polynomial plots. We then compared these outcomes among the groups using multivariable regression models, adjusting for child age, sex, mother's religion, and scheduled caste or tribe (designated groups of historically disadvantaged people in India). To determine the pathways through which adolescent pregnancy might contribute to child undernutrition, we used multivariable adjusted regression models to examine the association of adolescent pregnancy with the hypothesised linking factors. We used structural equation models, considering all the potential variables in our conceptual framework, to examine the direct and indirect links between adolescent pregnancy and child HAZ. Because sample sizes were smaller for some variables in the NFHS-4, three path models were run. Model 1, using the full sample, included variables for maternal nutritional status, education, access to health services, and living conditions. Model 2 included all variables in Model 1 plus the child feeding variables available only for the subsample of mothers with children 6–24 months of age. Model 3 included all variables in Model 1 plus the women's bargaining power variables in the subsample with available data. The indirect effects, calculated by multiplying coefficients for each path, allowed us to compare the relative strength of each path.

for the subsample of mothers with children 6–24 months of age. Model 3 included all variables in Model 1 plus the women's bargaining power variables in the subsample with available data. The indirect effects, calculated by multiplying coefficients for each path, allowed us to compare the relative strength of each path. As a robustness check for all models, we also added state fixed effects to account for state-level heterogeneity in externalities that might affect maternal and child health, such as state-specific programmes and policies. Role of the funding source The funder of the study had no role in study design, data collection, data analysis, data interpretation, or writing of the report. All authors had access to all the data and had final responsibility for the decision to submit. Results Of 60 096 women aged 15–49 included in the final sample (appendix), 14 107 (25·1%) first gave birth during adolescence, 31 475 (52·3%) during young adulthood, and 14 514 (23·2%) during adulthood. As expected, women who first gave birth during adolescence got married earlier (mean age at marriage 16·4 years of age, SD 1·67) compared with those who first gave birth as young adults (19·7, 2·14) or adults (24·3, 3·87; appendix). A greater proportion of women who first gave birth during adolescence lived in rural areas and belonged to a disadvantaged group compared with women who first gave birth as adults (appendix). The age and sex ratio of firstborn children was similar across groups (appendix).

riods had less of a role. Although adolescent pregnancy was associated with poorer access to services during the postpartum and early childhood periods, these indicators were not significantly associated with child HAZ in our multivariable regression model after controlling for maternal education and living conditions. Complementary feeding practices were identified as important in the relationship between adolescent pregnancy and child nutrition. In our study, children born to adolescent mothers were less likely to achieve adequate diet and consume iron-rich foods. Although the reasons for these findings are not possible to ascertain from the available survey data, poor complementary feeding among adolescent mothers could stem from poorer living conditions (which affect access to adequate diets), lower access to information through lower use of health services, both of which we find in our study, or a reduced cognitive or emotional ability to manage the demands of a young infant.24 Adequate diet has been found to have strong association with child growth, both from global25 and Indian data.26

(19·7, 2·14) or adults (24·3, 3·87; appendix). A greater proportion of women who first gave birth during adolescence lived in rural areas and belonged to a disadvantaged group compared with women who first gave birth as adults (appendix). The age and sex ratio of firstborn children was similar across groups (appendix). At all ages between 0 and 60 months, children born to adolescent mothers had lower HAZ and WAZ than children born to young adults or adult women (figure 2; appendix). For WHZ, a similar trend was observed in children older than 6 months, with slightly smaller group differences compared with HAZ and WAZ (appendix).Figure 2 HAZ score and prevalence of stunting for first-born children by child's age and mother's age at first birth in India, 2016 (A) HAZ score. Curves are smooth local polynomials with 95% CIs. (B) Stunting. Adjusted coefficient (95% CI) for models are shown below each panel in the figure. OLS regression models were adjusted for child age, sex, maternal religion, and caste fixed effects and controlled for the cluster sampling design and sampling weights used in the survey. Data are from India's fourth National Family Health Survey (NFHS-4, 2016).16 HAZ=Length or height-for-age Z score. OLS=ordinary least squares.

ssion models were adjusted for child age, sex, maternal religion, and caste fixed effects and controlled for the cluster sampling design and sampling weights used in the survey. Data are from India's fourth National Family Health Survey (NFHS-4, 2016).16 HAZ=Length or height-for-age Z score. OLS=ordinary least squares. The multivariable regression confirmed that children born to adolescent mothers were 0·25 SD (95% CI −0·31 to −0·18) shorter for their age on average and were 5 percentage points (95% CI 2–7) more likely to be stunted than children born to young adults (figure 2). These differences were more prominent when comparing children born to adolescents with those born to adults (β=–0·53 [95% CI −0·64 to −0·41] for height-for-age Z score and 11 percentage points [8–15] for stunting). Compared with children born to adult mothers, children born to adolescent mothers also had lower WAZ (β=–0·40, 95% CI −0·47 to −0·32), lower WHZ (β=–0·16, −0·23 to −0·08), and higher prevalence of underweight (10 percentage points, 7–12) but not wasting (appendix). Adolescent pregnancy was negatively associated with maternal nutritional status (table). Compared with women who first gave birth as adults, women who first gave birth during adolescence were shorter, weighed less, had lower haemoglobin, had lower body-mass index, and had a higher prevalence of underweight and anaemia (table).Table Association between maternal age at first birth and factors related to child nutritional status in India, 2016

birth as adults, women who first gave birth during adolescence were shorter, weighed less, had lower haemoglobin, had lower body-mass index, and had a higher prevalence of underweight and anaemia (table).Table Association between maternal age at first birth and factors related to child nutritional status in India, 2016 Total women First birth during adolescence (10–19 years) First birth during young adulthood (20–24 years) First birth during adulthood (≥25 years) Adolescence vs young adulthood* Adolescence vs adulthood* Number of women n (%) or mean Number of women n (%) or mean Number of women n (%) or mean β (95%CI) p value β (95%CI) p value Maternal nutritional status Height, cm 51 281 11 576 151·47 (5·54) 26 747 152·15 (5·82) 12 958 152·79 (6·12) −0·63 (−1·09 to −0·18) 0·0073 −1·21 (−1·78 to −0·65) <0·0001 Weight, kg 51 342 11 588 46·64 (8·27) 26 772 49·03 (9·52) 12 982 53·68 (11·34) −2·29 (−2·91 to 1·67) <0·0001 −6·81 (−7·71 to −5·91) <0·0001 BMI, kg/m2 51 242 11 568 20·30 (3·25) 26 725 21·15 (3·76) 12 949 22·96 (4·49) −0·82 (−1·02 to −0·62) <0·0001 −2·59 (−2·93 to −2·25) <0·0001 BMI <18·5 kg/m2 51 242 11 568 3777 (32·56%) 26 725 6581 (25·19%) 12 949 1756 (13·88%) 0·07 (0·05 to 0·09) <0·0001 0·18 (0·15 to 0·21) <0·0001 Haemoglobin, g/dL 51 076 11 520 11·55 (1·46) 26 663 11·65 (1·54) 12 893 11·83 (1·58) −0·09 (−0·14 to −0·04) 0·0010 −0·26 (−0·35 to −0·17) <0·0001 Anaemia 51 076 11 520 6512 (58·02%) 26 663 14 420 (54·83%) 12 893 6224 (48·97%) 0·03 (0·01 to 0·04) 0·0008 0·08 (0·06 to 0·11) <0·0001 Access to antenatal care services during pregnancy Early antenatal care 60 012 14 090 8296 (60·32%) 31 426 20 876 (67·54%) 14 496 10 494 (72·55%) −0·07 (−0·09 to −0·04) <0·0001 −0·12 (−0·16 to −0·07) <0·0001 At least 4 antenatal care visits 60 012 14 090 6893 (56·66%) 31 426 17 723 (60·57%) 14 496 9917 (72·05%) −0·04 (−0·11 to 0·03) 0·29 −0·15 (−0·26 to −0·04) 0·0068 Received iron and folic acid 59 920 14 059 11 122 (80·56%) 31 389 26 077 (83·73%) 14 472 12 605 (88·35%) −0·03 (−0·07 to 0·01) 0·098 −0·08 (−0·13 to −0·03) 0·0039 Received deworming 59 442 13 961 2253 (18·39%) 31 153 5509 (20·27%) 14 328 2797 (23·58%) −0·02 (−0·04 to 0·01) 0·14 −0·05 (−0·08 to −0·02) 0·0013 Access to delivery and postpartum services Institutional delivery 60 001 14 089 11 815 (86·95%) 31 419 28 058 (91·62%) 14 493 13 501 (95·61%) −0·04 (−0·05 to −0·03) <0·0001 −0·08 (−0·11 to −0·05) <0·0001 Skilled birth attendant at delivery 60 096 14 107 12 045 (88·28%) 31 475 28 377 (92·11%) 14 514 13 556 (95·18%) −

13 Access to delivery and postpartum services Institutional delivery 60 001 14 089 11 815 (86·95%) 31 419 28 058 (91·62%) 14 493 13 501 (95·61%) −0·04 (−0·05 to −0·03) <0·0001 −0·08 (−0·11 to −0·05) <0·0001 Skilled birth attendant at delivery 60 096 14 107 12 045 (88·28%) 31 475 28 377 (92·11%) 14 514 13 556 (95·18%) − 0·03 (−0·05 to −0·02) <0·0001 −0·06 (−0·09 to −0·04) <0·0001 Postnatal care for mothers 60 096 14 107 8756 (65·65%) 31 475 21 647 (70·80%) 14 514 11 002 (77·27%) −0·05 (−0·08 to −0·02) 0·0011 −0·11 (−0·16 to −0·06) <0·0001 Access to services during childhood Growth monitoring 60 096 14 107 6563 (50·43%) 31 475 14 319 (45·78%) 14 514 5518 (38·59%) 0·04 (−0·02 to 0·11) 0·21 0·11 (0·01 to 0·21) 0·025 Food supplementation 60 096 14 107 7863 (58·69%) 31 475 16 716 (52·05%) 14 514 6675 (43·35%) 0·06 (−0·01 to 0·12) 0·056 0·14 (0·05 to 0·23) 0·0025 Full immunisation† 41 667 9990 6082 (64·69%) 21 488 14 381 (67·21%) 10 189 7185 (70·74%) −0·02 (−0·08 to 0·03) 0·39 −0·06 (−0·13 to 0·00) 0·097 Vitamin A‡ 50 978 12 126 6886 (60·46%) 26 506 15 769 (62·14%) 12 346 7803 (66·68%) −0·02 (−0·05 to 0·02) 0·31 −0·06 (−0·08 to −0·04) <0·0001 Child iron and folic acid‡ 50 669 12 042 2945 (27·04%) 26 350 6978 (28·55%) 12 277 3424 (31·59%) −0·02 (−0·04 to 0·01) 0·19 −0·04 (−0·07 to −0·02) 0·0005 Child deworming‡ 50 647 12 030 3666 (35·26%) 26 351 8224 (33·76%) 12 266 4279 (37·61%) 0·01 (−0·04 to 0·07) 0·65 −0·02 (−0·09 to 0·06) 0·69 Infant and young child feeding practices Early initiation of breastfeeding§ 33 313 7831 3512 (43·59%) 17 999 7595 (42·49%) 7483 3352 (42·82%) 0·01 (−0·02 to 0·04) 0·60 0·01 (−0·03 to 0·03) 0·85 Exclusive breastfeeding¶ 8419 1777 1137 (63·38%) 4661 2845 (60·69%) 1981 1246 (59·50%) 0·03 (−0·01 to 0·07) 0·13 0·04 (−0·01 to 0·08) 0·10 Adequate diet‖ 25 494 6144 529 (8·69%) 13 673 1209 (9·13%) 5677 653 (11·15%) −0·01 (−0·02 to 0·01) 0·30 −0·03 (−0·05 to −0·01) 0·049 Consumption of iron-rich food‖ 25 871 6257 1430 (23·18%) 13 847 3251 (23·92%) 5767 1966 (32·75%) −0·01 (−0·04 to 0·02) 0·57 −0·09 (−0·13 to −0·06) <0·0001 Living conditions Socioeconomic status index 60 096 14 107 −0·32 (0·87) 31 475 0·03 (0·91) 14 514 0·39 (0·87) −0·33 (−0·41 to −0·25) <0·0001 −0·66 (−0·82 to −0·50) <0·0001 Improved sanitation facility 60 096 14 107 5509 (38·80%) 31 475 15 979 (51·93%) 14 514 9678 (68·58%) −0·12 (−0·16 to −0·09) <0·0001 −0·28 (−0·32 to −0·24) <0·0001 Education and bargaining power Education, years 60 096 14 107 7·27 (3·96) 31 475 8·96 (4·84) 14 514 10·81 (5·48) −1·5

<0·0001 −0·66 (−0·82 to −0·50) <0·0001 Improved sanitation facility 60 096 14 107 5509 (38·80%) 31 475 15 979 (51·93%) 14 514 9678 (68·58%) −0·12 (−0·16 to −0·09) <0·0001 −0·28 (−0·32 to −0·24) <0·0001 Education and bargaining power Education, years 60 096 14 107 7·27 (3·96) 31 475 8·96 (4·84) 14 514 10·81 (5·48) −1·5 4 (−1·83 to −1·26) <0·0001 −3·30 (−3·68 to −2·91) <0·0001 Works for pay 10 502 2348 411 (15·36%) 5409 852 (13·78%) 2745 598 (21·37%) 0·01 (−0·01 to 0·03) 0·37 −0·07 (−0·12 to −0·01) 0·015 Ownership of money 10 340 2302 755 (35·82%) 5341 2093 (40·43%) 2697 1329 (49·95%) −0·05 (−0·10 to 0·01) 0·084 −0·14 (−0·20 to −0·08) <0·0001 Ownership of land or house 10 561 2365 919 (32·85%) 5435 2072 (34·76%) 2761 1087 (36·59%) −0·02 (−0·06 to 0·03) 0·39 −0·04 (−0·11 to 0·03) 0·26 Say in household decisions 10 340 2302 0·66 (0·38) 5341 0·69 (0·39) 2697 0·74 (0·39) −0·03 (−0·06 to 0·00) 0·040 −0·08 (−0·12 to −0·04) 0·0007 Mobility 10 561 2365 0·38 (0·42) 5435 0·41 (0·45) 2761 0·53 (0·47) −0·03 (−0·11 to 0·04) 0·38 −0·15 (−0·24 to −0·07) 0·0008 Data are from India's fourth National Family Health Survey16 and adjusted to account for cluster sampling design and sampling weights. BMI=body-mass index. * Maternal outcomes adjusted for maternal caste and religion, and child outcomes adjusted for child age, gender, maternal caste, and religion fixed effects. † Children aged 12–59 months. ‡ Children aged >6 months. § Children aged 0–23 months. ¶ Children aged <6 months. ‖ Children aged 6–23 months.

4 (−1·83 to −1·26) <0·0001 −3·30 (−3·68 to −2·91) <0·0001 Works for pay 10 502 2348 411 (15·36%) 5409 852 (13·78%) 2745 598 (21·37%) 0·01 (−0·01 to 0·03) 0·37 −0·07 (−0·12 to −0·01) 0·015 Ownership of money 10 340 2302 755 (35·82%) 5341 2093 (40·43%) 2697 1329 (49·95%) −0·05 (−0·10 to 0·01) 0·084 −0·14 (−0·20 to −0·08) <0·0001 Ownership of land or house 10 561 2365 919 (32·85%) 5435 2072 (34·76%) 2761 1087 (36·59%) −0·02 (−0·06 to 0·03) 0·39 −0·04 (−0·11 to 0·03) 0·26 Say in household decisions 10 340 2302 0·66 (0·38) 5341 0·69 (0·39) 2697 0·74 (0·39) −0·03 (−0·06 to 0·00) 0·040 −0·08 (−0·12 to −0·04) 0·0007 Mobility 10 561 2365 0·38 (0·42) 5435 0·41 (0·45) 2761 0·53 (0·47) −0·03 (−0·11 to 0·04) 0·38 −0·15 (−0·24 to −0·07) 0·0008 Data are from India's fourth National Family Health Survey16 and adjusted to account for cluster sampling design and sampling weights. BMI=body-mass index. * Maternal outcomes adjusted for maternal caste and religion, and child outcomes adjusted for child age, gender, maternal caste, and religion fixed effects. † Children aged 12–59 months. ‡ Children aged >6 months. § Children aged 0–23 months. ¶ Children aged <6 months. ‖ Children aged 6–23 months. Women who first gave birth during adolescence had poorer access to and use of antenatal care services than did women who first gave birth during adulthood (table). Adolescent pregnancy was also associated with less service use during delivery and early postpartum and poorer access to services during early childhood (table). Adolescent pregnancy was positively associated with child growth monitoring and food supplementation (table), both interventions delivered by India's national Integrated Child Development Services programme.

s service use during delivery and early postpartum and poorer access to services during early childhood (table). Adolescent pregnancy was positively associated with child growth monitoring and food supplementation (table), both interventions delivered by India's national Integrated Child Development Services programme. Adolescent pregnancy was negatively associated with complementary feeding practices (table). No differences were found between adolescents, young adult, and adult mothers for breastfeeding. Adolescent pregnancy was negatively associated with optimal living conditions, women's education, and most bargaining power indicators except for work for pay and ownership of land or house. Compared with adult mothers, women who gave birth during adolescence were more likely to live in a household with lower socioeconomic status and poorer sanitation (table). Women who first gave birth during adolescence had 3·30 (95% CI −3·68 to −2·91) fewer years of education, were less likely to work for pay, had lower power in ownership of money, less say in household decisions, and less freedom of mobility (table).

ousehold with lower socioeconomic status and poorer sanitation (table). Women who first gave birth during adolescence had 3·30 (95% CI −3·68 to −2·91) fewer years of education, were less likely to work for pay, had lower power in ownership of money, less say in household decisions, and less freedom of mobility (table). In the full path model considering all available factors, we found evidence of strong links between adolescent pregnancy and child undernutrition through maternal nutritional status (proxied by height, weight, and haemoglobin), access to antenatal care services (at least four antenatal care visits and receiving iron and folic acid during pregnancy), women's education, living conditions (household socioeconomic status and sanitation), and adequate child diet (figure 3). The indirect path through these factors accounted for 68·5% of the relationship between adolescent pregnancy on child HAZ (ie, the 0·53 SD difference). Overall, after considering relative indirect effects for all pathways, the strongest links between adolescent pregnancy on child HAZ worked through education (which accounted for 18% of the diffence), socioeconomic status (13% of the difference), and maternal weight (15% of the difference).Figure 3 Pathways from adolescent pregnancy to child undernutrition through maternal nutrition, health services, infant and young child feeding practices, living conditions, women's education, and bargaining power

f the diffence), socioeconomic status (13% of the difference), and maternal weight (15% of the difference).Figure 3 Pathways from adolescent pregnancy to child undernutrition through maternal nutrition, health services, infant and young child feeding practices, living conditions, women's education, and bargaining power Bold font and shading for the mediating variables indicates that the variable was important in mediating the association between adolescent pregnancy and child height-for-age Z score (p<0·1 for both steps). Numbers in the paths are regression coefficients. Figure presents paths from the three path models that were run. Model 1 included variables for maternal nutritional status, access to health and nutrition services, living conditions, and women's education. Model 2 included all variables in model 1 plus the child feeding variables shown, in the subsample of mothers with children 6–24 months of age (n=25 494). Model 3 included all variables in model 1 plus the bargaining power variables shown, in the subsample with data available (n=10 340). All models adjusted for maternal caste and religion and child age and sex and controlled for the cluster sampling design and sampling weights used in the survey. Coefficients shown in the figure are from Model 1, except for those for child feeding (from Model 2) and bargaining power (from Model 3). HAZ=Length or height-for-age Z score. *Constructed from principal components analysis of asset ownership.

led for the cluster sampling design and sampling weights used in the survey. Coefficients shown in the figure are from Model 1, except for those for child feeding (from Model 2) and bargaining power (from Model 3). HAZ=Length or height-for-age Z score. *Constructed from principal components analysis of asset ownership. Discussion As the global adolescent population grows, reducing early childbearing is crucial to achieving the Sustainable Development Goals related to poverty, health, nutrition, general wellbeing, equity, and education. By use of data from India's largest nationally representative health survey in 2015–16, NFHS-4,16 we aimed to understand how early childbearing relates to child undernutrition through a set of social, biological, and programmatic factors related to adolescent pregnancy. We found that stunting and underweight were 11 percentage points more prevalent in children born to adolescent mothers compared with children born to adult mothers. Our path analyses showed that adolescent pregnancy is associated with poorer maternal nutritional status, lower educational attainment, lower likelihood of accessing antenatal health services, poorer complementary feeding practices, and poorer living conditions, all of which were also associated with child stunting.

s. Our path analyses showed that adolescent pregnancy is associated with poorer maternal nutritional status, lower educational attainment, lower likelihood of accessing antenatal health services, poorer complementary feeding practices, and poorer living conditions, all of which were also associated with child stunting. Compared with women who first gave birth as adults, women who were mothers in adolescence were shorter and thinner, and these maternal nutritional insults were associated with lower length or height and weight in their children. Although height and weight in our study were measured at the time of survey, up to 5 years after the mother's first birth, other evidence provides insights into the biological link between maternal and child anthropometry. For example, in a study done in Bangladesh, pregnancy and lactation ceased linear growth and resulted in weight loss and depletion of fat and lean body mass in adolescent girls (aged 12–19 years),12 probably due to concurrent competition for nutrients between the mother, who is still growing, and fetus. Similarly, a multinomial regression analysis using the third round of India's NFHS (2005–06) found a large adverse effect of early marriage and childbearing on thinness in women.13 In our study, adolescent mothers were more likely to be anaemic compared with adult mothers, and anaemia was associated with reduced child growth. However, these findings should be interpreted with caution as a mother's anaemia and weight status might not reflect her earlier status at the time of pregnancy. However, current maternal height could be more closely associated with height at the time of pregnancy, because adolescent height growth in girls usually peaks before menarche.

dings should be interpreted with caution as a mother's anaemia and weight status might not reflect her earlier status at the time of pregnancy. However, current maternal height could be more closely associated with height at the time of pregnancy, because adolescent height growth in girls usually peaks before menarche. Adolescent mothers had about 3 fewer years of education compared with adult mothers. They were also less likely to work for pay, to have money they could spend, to have a say in household decisions, and to be able to travel without permission. Among these factors, only education was related to child HAZ; the path was one of the strongest observed, with the effect of adolescent pregnancy on mother's education accounting for an 18% difference in child HAZ. School-aged girls (aged 11–17 years) who marry and bear children are more likely to discontinue their education compared with those who do not marry or become pregnant during school.21, 22 Furthermore, maternal education is known to be a strong determinant of many child outcomes, including linear growth.23 Women's bargaining power might still be an important determinant of child nutrition; however the effects might have been outweighed by education in our multivariable model.

o adequate diets), lower access to information through lower use of health services, both of which we find in our study, or a reduced cognitive or emotional ability to manage the demands of a young infant.24 Adequate diet has been found to have strong association with child growth, both from global25 and Indian data.26 Socioeconomic status and improved sanitation at the household level were key links between adolescent pregnancy and poor child nutrition. Adolescent pregnancy could cause poverty over multiple generations, but it is more likely that adolescent pregnancy would occur in high rather than low poverty contexts in the short term. Adolescent pregnancy might perpetuate the cycle of poverty, because women who bear children early are more likely to discontinue education and, thus, have lower earning potential. Previous published literature examining the consequences of adolescent childbearing has focused on direct mother and child effects but has not linked teen pregnancy to child nutrition outcomes through a wide set of factors faced by adolescents. Our study is the first to consider multiple pathways at multiple levels (social, biological, and programmatic factors) to explain the detrimental effects of early childbearing age on child undernutrition. Our findings are strengthened by the fact that we only included primiparous mothers. Although we sacrificed sample size using this approach, we felt that it was important given the strong confounding effect of child birth order and birth spacing on child nutritional status.15, 27

ildbearing age on child undernutrition. Our findings are strengthened by the fact that we only included primiparous mothers. Although we sacrificed sample size using this approach, we felt that it was important given the strong confounding effect of child birth order and birth spacing on child nutritional status.15, 27 Our study contains some methodological limitations that deserve consideration. The cross-sectional design reduces causal inference. For example, becoming pregnant early might lead to reduced education or wealth; however, a woman from a poor background and lower education might be more likely to become pregnant early. Thus, longitudinal data are needed to establish causal relationships. Recall bias in women with older children might have affected their responses to questions about prepregnancy health service use, but this bias is probably uniform across groups, because the child age distributions were similar for children born to adolescents, young adults, and adults. Current poor health and living conditions might reflect an equally poor situation for the mother in past years, and we accept that this history of combined insults probably leads to the child's current anthropometric state. Although women's bargaining power variables were only collected for a random subsample in NFHS-4, the characterisitics of this subsample was similar to the full sample (appendix). Therefore, we believe that the analysed sample provides unbiased estimates for the bargaining power pathway. We did not examine child marriage among boys, which was about the same as the percentage of girls married at less than 18 years of age.16 Future research should explore whether the effects of early childbearing on child undernutrition work through paternal pathways, especially in light of transitioning gender roles and efforts to shift the focus of paediatric care from the mother-child dyad to the mother-father-child triad.28

less than 18 years of age.16 Future research should explore whether the effects of early childbearing on child undernutrition work through paternal pathways, especially in light of transitioning gender roles and efforts to shift the focus of paediatric care from the mother-child dyad to the mother-father-child triad.28 Interventions to increase age at first marriage, age at first birth, and girl's education are a promising approach to also improve maternal and child nutrition. A review of interventions to prevent child marriage in low-income and middle-income countries found that six of 11 high-quality studies showed some positive effect on decreasing child marriage (marriage before 18 years of age) or increasing age at marriage—interventions included unconditional cash transfers, cash transfers conditional on school enrollment or attendance, school vouchers, life-skills curriculum, and livelihood training.29 In the past 25 years, the Government of India has piloted different cash transfers conditional on education, with complementary programming meant to encourage investment in girls’ human capital, and several adolescent health programmes are ongoing under different ministries in India.30 We acknowledge the strong cultural barriers to efforts to delaying early marriage in this context; however, the declines already seen in the past decade suggest that further investments in appropriate interventions targeting young people, including men, can contribute to further declines in early marriage and early childbearing in India. The high variability in adolescent pregnancy across states and districts suggests that subnational policies and programmes targeting early marriage and early childbearing might well be needed to address differences in cultural practices and other conditions affecting early marriage and early childbearing.

ing in India. The high variability in adolescent pregnancy across states and districts suggests that subnational policies and programmes targeting early marriage and early childbearing might well be needed to address differences in cultural practices and other conditions affecting early marriage and early childbearing. Investments in adolescents have a high benefit-to-cost ratio and yield a triple dividend on the health and wellbeing of adolescents, adults, and the next generation.2 Early marriage and childbearing are at the root of many other adolescent issues, because these practices have adverse effects on health, education, and future employment, all of which have intergenerational consequences. The evidence presented here suggests that adolescent pregnancy also has substantial consequences for India's undernutrition burden, operating through multiple factors. Continued investments, especially if focused in areas and communities with high levels of early marriage and childbearing, could have long-ranging positive effects. Supplementary Material Supplementary appendix Acknowledgments This study was funded by the Bill & Melinda Gates Foundation through Partnerships and Opportunities to Strengthen and Harmonize Actions for Nutrition in India, led by International Food Policy Research Institute.

Investments in adolescents have a high benefit-to-cost ratio and yield a triple dividend on the health and wellbeing of adolescents, adults, and the next generation.2 Early marriage and childbearing are at the root of many other adolescent issues, because these practices have adverse effects on health, education, and future employment, all of which have intergenerational consequences. The evidence presented here suggests that adolescent pregnancy also has substantial consequences for India's undernutrition burden, operating through multiple factors. Continued investments, especially if focused in areas and communities with high levels of early marriage and childbearing, could have long-ranging positive effects. Supplementary Material Supplementary appendix Acknowledgments This study was funded by the Bill & Melinda Gates Foundation through Partnerships and Opportunities to Strengthen and Harmonize Actions for Nutrition in India, led by International Food Policy Research Institute. Contributors PHN and SS conceived the idea for the manuscript and wrote substantial parts. PHN led the overall synthesis of the manuscript. SS and SN did the literature review. PHN, SN, and LMT did the statistical analysis, which was reviewed by SS and PM. LMT prepared the table and figures. PM supported data interpretation and reviewed and edited the manuscript. All authors read and approved the final submitted manuscript. Declaration of interests We declare no competing interests.

* OMA evaluation time points include: at Dx, monthly for 6 months, then every other month until 1 year, 18 months, 2 years and yearly thereafter two months, six months and one year. Figure 2 CONSORT Diagram for COG Study ANBL00P3 Figure 3 a. Neuroblastoma event-free, OMA progression-free, and overall survival of the overall patient cohort (n=53) b. OMA progression-free survival for IVIG+ (n=26) versus NO-IVIG (n=27) Table 1 Opsoclonus Myoclonus Syndrome severity scale, by Drs. Wendy G. Mitchell and Michael G. Pike (5)

Introduction Opsoclonus myoclonus ataxia syndrome (OMA), also known as dancing eyes and dancing feet syndrome or Kingsbourne syndrome,(1) is a rare neurologic disorder that affects 2–3% of the > 650 children diagnosed with neuroblastoma every year in North America.(2) Symptoms include conjugate rapid eye movements; spontaneous muscle jerking which can affect the trunk, face and extremities; ataxia; personality changes including irritability and behavior disorders; and developmental regression. OMA also occurs in children and adults without the diagnosis of neuroblastoma and may be triggered by intercurrent infection, but in many subjects the triggering event is never identified.(3) The proportion of children with neuroblastoma-associated OMA varies according to the cohort analyzed.(1) In a retrospective study of patients treated at two large pediatric oncology programs and two large neurology centers in France, 22 (64%) of 34 children with OMA had associated neuroblastoma.(4) The majority of children with neuroblastoma-associated OMA have low-risk neuroblastoma and are cured of their neuroblastoma with surgery alone or surgery with moderate-dose chemotherapy.(3–6) However, the neurological sequelae of OMA are often severe and lifelong.(5;6)

children with OMA had associated neuroblastoma.(4) The majority of children with neuroblastoma-associated OMA have low-risk neuroblastoma and are cured of their neuroblastoma with surgery alone or surgery with moderate-dose chemotherapy.(3–6) However, the neurological sequelae of OMA are often severe and lifelong.(5;6) Although the cause of OMA remains unknown, there is significant evidence that the disorder results from an autoimmune process. Serum autoantibodies against neuronal tissues have been identified in some patients with neuroblastoma-associated OMA.(7;8) Several groups have documented the presence of B-cells in the cerebrospinal fluid, increased B-cell activating factor in serum and cerebrospinal fluid, and other B-cell related cytokines and increased tumor infiltrating lymphocytes, both B- and T-cells.(9–12) However, the most compelling evidence for the autoimmune nature of this disorder is the clinical response to corticosteroids, intravenous gamma globuilin, rituximab, and/or other immunosuppressive therapy reported in single cases or small retrospective series. (1;13;14) Further, a retrospective analysis of 29 children with neuroblastoma and OMA from the Pediatric Oncology Group (POG) indicated that the immune suppression associated with chemotherapy may also be beneficial to patients with neuroblastoma-associated OMA.(6) All ten children in this series who received chemotherapy as part of their neuroblastoma treatment had resolution of their acute OMA symptoms and six had no long-term neurologic sequelae.

ed that the immune suppression associated with chemotherapy may also be beneficial to patients with neuroblastoma-associated OMA.(6) All ten children in this series who received chemotherapy as part of their neuroblastoma treatment had resolution of their acute OMA symptoms and six had no long-term neurologic sequelae. Because of the rarity of this condition, no previous prospective clnical trials have been conducted, and published retrospective series include only small numbers of patients. Thus, the expected OMA response rate to corticosteroids alone or combination immunosuppressive regimens is not known. Based on the promising responses to chemotherapy reported in the retrospective analysis of POG patients,(6) we hypothesized that immunosuppressive therapy with prednisone plus risk-adpated chemotherapy (with cyclophosphamide for low-risk patients) would alleviate the acute neurologic symptoms of OMA and also improve the long-term neurologic outcome. We further hypothesized that the addition of IVIG, an immune modulatory agent, would augment the neurologic recovery in these patients.(1) To test these hypotheses, the Children’s Oncology Group (COG) conducted a prospective randomized phase III clinical trial (ANBL00P3) for children with neuroblastoma-associated OMA, with a primary endpoint of OMA response.

n of IVIG, an immune modulatory agent, would augment the neurologic recovery in these patients.(1) To test these hypotheses, the Children’s Oncology Group (COG) conducted a prospective randomized phase III clinical trial (ANBL00P3) for children with neuroblastoma-associated OMA, with a primary endpoint of OMA response. Patients and Methods Study Design This trial was approved by the COG and made available to the more than 200 COG institutions. Ninety-two of these institutions opened the trial for enrolment. The study design is a randomized open label clinical trial. This is a standard approach for children with malignancies when the treatment is intravenous and it is impractical across a large cooperative group like COG and unethical to blind the investigators and expose children to an intravenous placebo arm.

enrolment. The study design is a randomized open label clinical trial. This is a standard approach for children with malignancies when the treatment is intravenous and it is impractical across a large cooperative group like COG and unethical to blind the investigators and expose children to an intravenous placebo arm. Participants Subjects ≤8 years of age with biopsy-proven, newly diagnosed neuroblastoma or ganglioneuroblastoma associated with OMA were eligible for the study. Any one of the neurologic components of OMA was sufficient for eligibility (Table 1). Patients with neuroblastoma diagnosed within 6 months of the development of OMA and patients with OMA diagnosed within six months of the diagnosis of neuroblastoma were eligible. Patients could not have received prior chemotherapy. Patients who received treatment with prednisone or adrenocorticotropic hormone (ACTH) for ≤14 days were eligible. Patients had to be registered on study no later than 4 weeks after they were deemed to be eligible for the study. Other requirements were that patients had to be free of any organ dysfunction or disorder that would preclude the use of corticosteroid therapy, cyclophosphamide or risk-adapted chemotherapy, and IVIG. Staging of disease was according to the International Neuroblastoma Staging System (INSS).(15) Tumor histopathology, MYCN status, and DNA index were determined in central COG laboratories. The COG Tracking Center assigned neuroblastoma risk-group classification, as described previously.(15) Before the start of therapy, institutional review board approval at participating sites was obtained. Written informed consent was obtained before enrollment onto ANBL00P3 and the neuroblastoma biology study, ANBL00B1.

he COG Tracking Center assigned neuroblastoma risk-group classification, as described previously.(15) Before the start of therapy, institutional review board approval at participating sites was obtained. Written informed consent was obtained before enrollment onto ANBL00P3 and the neuroblastoma biology study, ANBL00B1. Randomization and masking Subjects were randomized at the COG statistical office and institutions informed of the unmasked randomization assignment. Randomization was stratified according to the neuroblastoma clinical stage.

he COG Tracking Center assigned neuroblastoma risk-group classification, as described previously.(15) Before the start of therapy, institutional review board approval at participating sites was obtained. Written informed consent was obtained before enrollment onto ANBL00P3 and the neuroblastoma biology study, ANBL00B1. Randomization and masking Subjects were randomized at the COG statistical office and institutions informed of the unmasked randomization assignment. Randomization was stratified according to the neuroblastoma clinical stage. Procedures Treatment All patients with low-risk neuroblastoma received prednisone and cyclophosphamide and were randomly assigned to also receive: i) IVIG (IVIG+); or ii) no IVIG (NO-IVIG). Patients with intermediate- or high-risk disease were treated with prednisone and corresponding risk-adapted therapy regimens,(16;17) although three intermediate-risk patients received prednisone and cyclophosphamide. The intermediate and high-risk patients were also randomized to receive IVIG or no additional therapy (Figure 1). For low-risk patients, treatment cycles were repeated every 28 days. Prednisone was given at a dose of 2 mg/kg per day divided twice a day for a minimum of two cycles if a complete response was achieved at the two month evaluation. A slow taper of the steroid dose could then be started and titrated to the clinical response. Prednisone could be continued until achieving a response for as long as 18 months, with a slow taper to maintain maximal clinical response. Subjects on NO-IVIG regimen that failed to respond or progressed any time after completing the two month evaluation were allowed to cross-over to the IVIG+ regimen but considered treatment non-responders upon crossing over (see statistical section). Subjects on the IVIG+ who failed to respond or progressed any time after the two month evaluation were allowed to switch to ACTH therapy instead of prednisone but were considered treatment non-responders upon switching (see statistical section). Patients received IV cyclophosphamide at a dose of 25 mg/kg for children ≤20 kg or 750 mg/m2 for children >20 kg on day 0 of each cycle for 6 cycles. Patients randomized to receive IVIG received 1 gm/kg on day 0 and 1 of cycle one; day 0 of cycles 2 to 6; and then on day 0 of cycles 8, 10 and 12. Based on the review of the literature available at the time of study design, which consisted of case reports and small series, one year of therapy was frequently reported. We, thus, empirically chose one year of therapy for this trial.

nd 1 of cycle one; day 0 of cycles 2 to 6; and then on day 0 of cycles 8, 10 and 12. Based on the review of the literature available at the time of study design, which consisted of case reports and small series, one year of therapy was frequently reported. We, thus, empirically chose one year of therapy for this trial. The protocol did not specify an immunization administration policy because it is a well-accepted practice in the pediatric oncology community to withhold immunization in this patient population.

Figure 2 CONSORT Diagram for COG Study ANBL00P3 Figure 3 a. Neuroblastoma event-free, OMA progression-free, and overall survival of the overall patient cohort (n=53) b. OMA progression-free survival for IVIG+ (n=26) versus NO-IVIG (n=27) Table 1 Opsoclonus Myoclonus Syndrome severity scale, by Drs. Wendy G. Mitchell and Michael G. Pike (5) Stance: 0: standing and sitting balance normal for age 1: mildly unstable standing for age, slightly wide-base 2: unable to stand without support but can sit without support 3: unable to sit without using hands to prop or other support Gait: 0: walking normal for age 1: mild wide-based gait for age but able to walk indoors and outdoors independently 2: walks only or predominantly with support from person or equipment 3: unable to walk even with support from person or equipment Arm and hand function: 0: normal for age 1: mild infrequent tremor or jerkiness without functional impairment 2: fine motor function (e.g. pincer grip of small objects, pencil use) persistently impaired for age but less precise manipulative tasks (e.g. playing with larger toys, feeding, dressing) normal or almost normal 3: major difficulty with all age-appropriate manipulative tasks Opsoclonus: 0: none 1: rare or only when elicited by change in fixation 2: frequent; interfering frequently with fixation and/or tracking 3: persistent; interfering continuously with fixation and tracking Mood/behavior 0: normal 1: mild increase in irritability but consolable and/or mild sleep disturbance but easily settled 2: irritability and sleep disturbance, interfering substantially with child and family life 3: persistent severe distress For all children aged 17 months or less, the following abilities were considered when making the determination of the OMA ratings: Able to hold head consistently erect when trunk vertical? Able to reach and grasp objects with each hand? Able to roll back to front and front to back? Able to finger-feed self? Table 2 Characteristics of 53 patients with OMA on COG study ANBL00P3

nd 1 of cycle one; day 0 of cycles 2 to 6; and then on day 0 of cycles 8, 10 and 12. Based on the review of the literature available at the time of study design, which consisted of case reports and small series, one year of therapy was frequently reported. We, thus, empirically chose one year of therapy for this trial. The protocol did not specify an immunization administration policy because it is a well-accepted practice in the pediatric oncology community to withhold immunization in this patient population. Outcomes OMA Response Evaluation The OMA symptoms were evaluated using the previously published neurologic severity scale of Mitchell and Pike (Table 1).(5) Using this system we evaluated stance, gait, arm and hand function, opsoclonus, and mood/behavior at enrollment. Each symptom was scored as 0 (no symptoms); 1 (mild symptoms); 2 (moderate symptoms); or 3 (severe symptoms). The patient’s neurologic symptoms were evaluated and scored at two months, six months, and one year after enrollment for assessment of the primary endpoint of OMA response (Supplemental Table 1). Supplemental Table 2 provides the average and distribution of the total and subscale OMA severity scores at two, six and 12 months. Response was defined as the best score of the three evaluations points. For the secondary endpoint of OMA progression-free survival (OMA-PFS), subjects were followed at their local institutions and the status of OMA neurologic symptoms listed in Table 1 were reported annually for up to ten years. For each symptom, the OMA response was classified as complete response (CR), partial response (PR), no response (NR), or progressive/worsening of OMA (PD). CR was defined as improvement from baseline OMA to normal at two-months, six-months, or one-year without progression of neurologic symptoms; a PR was defined as improvement from baseline OMA at two-months, six-months, or one-year without progression; NR was defined as no change in neurologic symptoms from baseline at two-months, six-months, and one-year; and PD was defined as sustained worsening of neurologic symptoms from baseline at two-months, six months, or one-year in any symptom. For the purposes of the statistical efficacy rule, a patient was classified as an overall OMA “responder” if the combination of the symptom-specific responses were either a CR or a PR (Supplemental Table 1). Subjects were otherwise classified as an overall OMA “non-responder”. All patients who crossed over from NO-IVIG to IVIG+, or switched to ACTH at any time, were classified as overall OMA non-responders.

OMA “responder” if the combination of the symptom-specific responses were either a CR or a PR (Supplemental Table 1). Subjects were otherwise classified as an overall OMA “non-responder”. All patients who crossed over from NO-IVIG to IVIG+, or switched to ACTH at any time, were classified as overall OMA non-responders. Neuroblastoma Response Evaluation Eligible intermediate- and high-risk patients were evaluated for neuroblastoma response using criteria specified in the intermediate- or high-risk treatment protocols. Evaluation of tumor response for low-risk patients treated with cyclophosphamide was conducted according to the guidelines of the COG low-risk neuroblastoma protocol P9641.(18) Secondary Outcomes One of the secondary outcomes predefined in our trial protocol was to better define neuroblastoma outcome in children with OMA, and these data are included in this report. Additional secondary outcomes were: 1) to determine if IVIG improved the functional outcome of OMA; 2) to determine the long-term neuro-psychological prognosis of children with OMA; and 3) to better define the epidemiology and pathogenesis of OMA by analyzing serum and cerebrospinal fluid samples for specific anti-neuronal antibodies and evaluating brain magnetic resonance imaging. These outcomes are currently being analyzed and will be included in future reports.

l prognosis of children with OMA; and 3) to better define the epidemiology and pathogenesis of OMA by analyzing serum and cerebrospinal fluid samples for specific anti-neuronal antibodies and evaluating brain magnetic resonance imaging. These outcomes are currently being analyzed and will be included in future reports. Statistical Analysis The randomization was stratified by risk group (low, intermediate, high) according to the COG Risk Classification System.(15;19) Analyses were conducted as intent-to-treat, except for a secondary analysis of the primary objective to assess the OMA response within evaluable patients, defined as those who completed the baseline evaluation and at least one other evaluation of OMA response at 2, 6, or 12 months. An interim futility monitoring rule was used to determine that a treatment arm met a minimum required response rate; arm(s) not meeting the minimum were eliminated (see Supplemental Statistical Methods). If both arms met the minimum response rate, a test of proportions was used to identify the arm with the superior OMA response rate, with p<0.2 considered statistically significant. Due to the rarity of this disease, and because it was deemed unlikely that the addition of IVIG would result in a lower reponse rate, the risk of a higher than usual type 1 error was considered acceptable. With a planned sample size of 26 evaluable patients per arm (52 total), the test of proportions had 90% power to detect a 28% difference and 80% power to detect a 23% difference in the proportion of OMA responders for NO-IVIG versus IVIG+.

se rate, the risk of a higher than usual type 1 error was considered acceptable. With a planned sample size of 26 evaluable patients per arm (52 total), the test of proportions had 90% power to detect a 28% difference and 80% power to detect a 23% difference in the proportion of OMA responders for NO-IVIG versus IVIG+. For event-free survival (EFS), time to event was calculated from study enrollment until the first occurrence of neuroblastoma relapse or progression, secondary malignancy, or death from any cause; patients without an event were censored at the date of last contact. For overall survival (OS), time to event was calculated from study enrollment until death from any cause; living patients were censored at the date of last contact. The Data Monitoring Committee for the protocol recommended that we add OMA-PFS as a secondary analysis of the results. For OMA-PFS, time to event was calculated from study enrollment until time of OMA progression, OMA non-response (see OMA response evaluation above), crossover to IVIG, or switch to ACTH; otherwise a patient was censored at the date last known to be OMA progression-free. Survival curves were generated using the methods of Kaplan and Meier,(20) with standard errors per Peto et al.(21) EFS and OS rates are presented as survival estimates (95% confidence interval (CI)). Within the subset of patients who crossed over to IVIG or switched to ACTH, OMA responses after the crossover or switch were descriptively summarized

d using the methods of Kaplan and Meier,(20) with standard errors per Peto et al.(21) EFS and OS rates are presented as survival estimates (95% confidence interval (CI)). Within the subset of patients who crossed over to IVIG or switched to ACTH, OMA responses after the crossover or switch were descriptively summarized Role of the Funding Source This study is supported by the Chair’s Grant U10 CA-98543 and CA-180886 and Statistical and Data Center Grant U10 CA-98413 and CA-180899 of the Children’s Oncology Group from the National Cancer Institute, National Institutes of Health, Bethesda, MD. The sponsoring institution does not have a role in study design, data collection, analysis or interpretation or role in writing this report. The corresponding author and statistician co-authors had full access to the data and all authors participated in the designs, analysis, interpretation of the data and the writing of the manuscript. This study is registered with Clinical Trials.gov (identifier NCT00033293).

interpretation or role in writing this report. The corresponding author and statistician co-authors had full access to the data and all authors participated in the designs, analysis, interpretation of the data and the writing of the manuscript. This study is registered with Clinical Trials.gov (identifier NCT00033293). Results Patients A total of 53 eligible patients, from the 92 institutions that opened the trial, were enrolled after the study was activated on May 15, 2004. The study closed to accrual on February 4, 2013. All 53 patients were included in the intent-to-treat analysis. However, one patient was not evaluable for the OMA response analysis because the patient was not seen in the required follow-up visit for neurologic assessment. This patient refused therapy post-randomization but prior to start of therapy; therefore, 52 patients were evaluable for toxicity [graded per Common Terminology Criteria for Adverse Events (CTCAE) version 4.0] and for the secondary analysis of OMA response. The median age at diagnosis was 18.9 (IQR: 16.6, 24.0) months (range 2.9–49.2 months). Thirty-three patients were female and 20 were male with a female:male ratio of 1.7:1.0 (Table 2). Forty-four patients had low-risk neuroblastoma, seven had intermediate-risk neuroblastoma, and two were diagnosed with MYCN-amplified, high-risk neuroblastoma according to the COG Risk Classification System.(15;19) Sixty-four percent of patients (n=34) had favorable histology according to the International Pathology Classification System (INPC) criteria.(22)

astoma, seven had intermediate-risk neuroblastoma, and two were diagnosed with MYCN-amplified, high-risk neuroblastoma according to the COG Risk Classification System.(15;19) Sixty-four percent of patients (n=34) had favorable histology according to the International Pathology Classification System (INPC) criteria.(22) Twenty-six subjects were randomized to IVIG+ and 27 to NO-IVIG. Four of the seven subjects with intermediate-risk neuroblastoma and both high-risk patients were treated with risk-based multi-agent chemotherapy. Thus, a total of 47 (44 low risk and 3 intermediate risk) patients were treated with cyclophosphamide and prednisone with or without IVIG (Figure 2). OMA Response Applying the protocol futility interim monitoring rule to test the lack of efficacy, neither treatment arm was eliminated for insufficient OMA response rate (see Supplemental Statistical Methods). The OMA response rate for IVIG+ (21/26=80.8%) was statistically significantly higher than for NO-IVIG (11/27=40.7%) (odds ratio=6.1; 95% CI: (1.5, 25.9), p=0.0029; Table 3). In a secondary analysis of 52 patients who received therapy and completed at least one follow-up assessment for OMA response, the OMA response rate remained statistically significantly higher for the IVIG+ cohort compared to the NO-IVIG cohort (p=0.0044).

40.7%) (odds ratio=6.1; 95% CI: (1.5, 25.9), p=0.0029; Table 3). In a secondary analysis of 52 patients who received therapy and completed at least one follow-up assessment for OMA response, the OMA response rate remained statistically significantly higher for the IVIG+ cohort compared to the NO-IVIG cohort (p=0.0044). Twelve patients randomized to NO-IVIG crossed over and received IVIG+ at a median time of 143.5 (IQR: 92, 251; range 1–372) days (Table 3). These patients were classified as non-responders. OMA symptoms did not improve and prednisone was discontinued in three patients randomized to IVIG+ and three patients randomized to NO-IVIG. All six patients were treated with ACTH and classified as non-responders. The 3-year OMA-PFS was 39.5% (25.3%, 53.7%) for the overall cohort (Figure 3a). The 27 patients randomized to IVIG+ had a higher OMA-PFS compared to the 26 patients treated with NO-IVIG [53.9% (34.0%, 73.7%) versus 24.2% (5.7%, 42.6%) respectively] (Figure 3b). The median follow-up time for patients who were OMA progression-free was 6.1 (IQR: 3.6, 8.7; range: 1 day, 10.3) years [6.2 (IQR: 4.2, 6.7; range: 1.6, 9.8) years for IVIG+, 4.6 (IQR: 1.0, 9.5; range: 1 day, 10.3) years for NO-IVIG].

th NO-IVIG [53.9% (34.0%, 73.7%) versus 24.2% (5.7%, 42.6%) respectively] (Figure 3b). The median follow-up time for patients who were OMA progression-free was 6.1 (IQR: 3.6, 8.7; range: 1 day, 10.3) years [6.2 (IQR: 4.2, 6.7; range: 1.6, 9.8) years for IVIG+, 4.6 (IQR: 1.0, 9.5; range: 1 day, 10.3) years for NO-IVIG]. Toxicity The combination of risk-based chemotherapy with or without IVIG was generally well tolerated Five of the low-risk patients developed fever, and one of these patients was admitted to the hospital. A total of twenty-eight subjects reported a Grade 3 or higher toxicity. Toxicity for patients with intermediate- and high-risk neuroblastoma who received multi-agent cytotoxic chemotherapy consisted predominantly of the expected hematological toxicities (chemotherapy induced neutropenia and thrombocytopenia). One patient with high-risk neuroblastoma developed overwhelming adenovirus infection after high-dose chemotherapy followed by autologous stem cell reconstitution and died of the infection. Table 4 summarizes the toxicities that were reported by institutions for each of the patients enrolled on this study. Because the patients randomized to the IVIG+ arm also received corticosteroids and chemotherapy it is not possible to determine if any of the toxicities were due specifically to the IVIG. It is likely that the reported nervous toxicties were related to the patients’ underlying OMA syndrome rather than the treatment received.

e the patients randomized to the IVIG+ arm also received corticosteroids and chemotherapy it is not possible to determine if any of the toxicities were due specifically to the IVIG. It is likely that the reported nervous toxicties were related to the patients’ underlying OMA syndrome rather than the treatment received. Neuroblastoma Survival The neuroblastoma 3-year EFS and OS for the entire cohort were 94.1% (87.3%, 100%) and 98.0% (94.1%, 100%), respectively (n=53, Figure 3a). The median follow-up time for patients without a neuroblastoma event was 6.0 (IQR: 3.8, 6.7; range: 1 day, 10.3) years [6.1 (IQR: 4.1, 6.7; range: 1.6, 9.8) years for IVIG+, 5.0 (IQR: 3.2, 7.5; range: 1 day, 10.3) years for NO-IVIG]. For low-risk patients, the 3-year EFS and OS were 97.6% (92.9%, 100%) and 100%, respectively. The median follow-up time for low-risk patients without an event was 5.9 (IQR: 3.8, 6.7; range: 1 day, 10.3) years [6.0 (IQR: 4.1, 6.7; range: 1.6, 9.8) years for IVIG+, 5.6 (IQR: 3.2, 6.6; range: 1 day, 10.3) years for NO-IVIG]. Of the seven intermediate-risk patients, six are alive without evidence of neuroblastoma disease progression. One intermediate-risk patient developed neuroblastoma disease progression associated with OMA relapse, but remains alive. One of the 2 high-risk patients is alive without evidence of neuroblastoma disease progression.

seven intermediate-risk patients, six are alive without evidence of neuroblastoma disease progression. One intermediate-risk patient developed neuroblastoma disease progression associated with OMA relapse, but remains alive. One of the 2 high-risk patients is alive without evidence of neuroblastoma disease progression. Discussion The acute neurologic symptoms associated with OMA are often severe, and long-term neurological deficits are common.(5;6) To date, treatment regimens have been based on case reports and small retrospective studies. In an effort to improve neurologic recovery in this patient cohort, we designed and conducted the first prospective, randomized clinical trial for children with neuroblastoma-associated OMA. This neurologic disorder appears to be caused by an autoimmune process, and corticosteroids have been widely used to treat OMA.(1) Prednisone was, therefore, included in both treatment arms in this clinical trial. Based on promising results of a retrospective analysis of neuroblastoma patients with OMA that suggested chemotherapy may enhance the resolution of both acute and long-term neurology sequelae (6) all patients in this study received chemotherapy. Low-risk patients received cyclophosphamide at doses frequently used for immunosuppression. Three intermediate-risk patients were also treated with cyclophosphamide, while the other 4 intermediate-risk patients and the 2 high-risk patients received risk-adpated treatment regimens.(15) Because chemotherapy was included in both treatment arms, it is not possible to assess the clinical impact of chemotherapy on OMA response.

diate-risk patients were also treated with cyclophosphamide, while the other 4 intermediate-risk patients and the 2 high-risk patients received risk-adpated treatment regimens.(15) Because chemotherapy was included in both treatment arms, it is not possible to assess the clinical impact of chemotherapy on OMA response. IVIG is also known to modulate the immune response in other autoimmune diseases, and case reports have indicated clinical response to this agent in patients with neuroblastoma-associated OMA.(23;24) To test if the addition of IVIG to prednisone and chemotherapy improved neurologic recovery, this study randomized patients to receive IVIG (IVIG+) or no IVIG (NO-IVIG). We found that patients randomized to IVIG+ had a statistically significantly higher rate of resolution of OMA symptoms compared to those who were randomized to NO-IVIG (p=0.0029). This study was not conducted in a blinded fashion due to ethical and practical concerns regarding intravenous infusion of placebo in pediatric patients. However, because pediatric oncologists routinely enroll patients on unblinded randomized cooperative group trials and remain in equipoise regarding the treatment arms, the vast majority of patients are treated as randomized.

ethical and practical concerns regarding intravenous infusion of placebo in pediatric patients. However, because pediatric oncologists routinely enroll patients on unblinded randomized cooperative group trials and remain in equipoise regarding the treatment arms, the vast majority of patients are treated as randomized. In contrast to the slight predominance of males observed in most neuroblastoma cohorts,(25) 62% (33/53) of the neuroblastoma patients with OMA in our study were female. Although female preponderance is a finding that is common in autoimmune disorders, this particular characteristic of neuroblastoma associated OMA patients has not been recognized previously. Our review of the published literature confirmed a preponderance of female cases in other OMA reports (3–5;26–28).

y were female. Although female preponderance is a finding that is common in autoimmune disorders, this particular characteristic of neuroblastoma associated OMA patients has not been recognized previously. Our review of the published literature confirmed a preponderance of female cases in other OMA reports (3–5;26–28). The majority of children with OMA and neuroblastoma have been reported to have low-risk disease,(1) and our study confirms the association between OMA and favorable biologic features. Forty-four (83%) of the 53 patients enrolled on the study had low-risk disease and seven had intermediate-risk. Only two patients were classified as high-risk. Reflecting the high proportion of low-risk tumors, survival of this cohort was excellent, with 3-year EFS and OS rates of 94.1% (87.3%, 100%) and 98.0% (94.1%, 100%), respectively. The single patient enrolled on this study who died had high-risk neuroblastoma and suffered an infectious complication following myeloablative therapy. Interestingly, the immunoglobulin fraction of serum from OMA patients has been shown to suppress growth of neuroblastoma cell lines,(29) suggesting there may be a biologic rationale for the low numbers of high-risk patients with OMA. However, it is well recognized that the majority of patients with OMA and neuroblastoma have significant neurologic morbidities, and many experience life-long neurologic deficits.(5;6) Additional follow-up is needed to assess the incidence of long-term neurologic sequelae in our cohort, which may include developmental delay and learning disabilities.

hat the majority of patients with OMA and neuroblastoma have significant neurologic morbidities, and many experience life-long neurologic deficits.(5;6) Additional follow-up is needed to assess the incidence of long-term neurologic sequelae in our cohort, which may include developmental delay and learning disabilities. OMA symptoms may be exacerbated by concurrent infection. We did not detect any difference in the incidence of infections rising to Grade 3 toxicity or greater infections in either treatment group. Because we did not collect data on lower toxicity grade infections, we are not able to determine if viral infections impacted OMA response rate.

xacerbated by concurrent infection. We did not detect any difference in the incidence of infections rising to Grade 3 toxicity or greater infections in either treatment group. Because we did not collect data on lower toxicity grade infections, we are not able to determine if viral infections impacted OMA response rate. Mitchell and co-workers, in their recent long-term follow-up report of OMA, showed a correlation of outcome and intensity of therapy consistent with the hypothesis that more intensive immunosuppression may be associated with improved long-term neurologic outcome.(30) Other studies have also indicated that a more aggressive treatment approach may lead to improved neurologic outcome for patients with OMA.(5;13) Our study demonstrates the addition of IVIG to prednisone and risk-adapted chemotherapy improves OMA response, indicating that IVIG+ will serve as an effective treatment back-bone to test future strategies. We conducted the trial within the COG which consists of over 200 institutions of varying sizes in North America, Australia and New Zealand. Most, if not all, children diagnosed with neuroblastoma are referred to a COG institution in North America. Although only 92 COG institutions elected to open the trial, and only a subset of these 92 institutions enrolled at least one eligible subject, this study was successfully conducted in small institutions as well as medium-sized hospitals and tertiary centers. These results suggest that the care and complex clinical evaluations required to treat this population of patients can be successfully applied in different hospital settings and are generalizable. At the Advances in Neuroblastoma Research meeting in Genoa, Italy 2004, a task force recommended that three of four of the following criteria be present to diagnose neuroblastoma with OMA: 1) opsoclonus, 2) myoclonus/ataxia, 3) behavioral changes and/or sleep disturbances, and 4) neuroblastoma. Although we did not apply the Genoa criteria in our study, only one of our patients (randomized to NO-IVIG) had less than 3 of the 4 Genoa criteria.(1)

owing criteria be present to diagnose neuroblastoma with OMA: 1) opsoclonus, 2) myoclonus/ataxia, 3) behavioral changes and/or sleep disturbances, and 4) neuroblastoma. Although we did not apply the Genoa criteria in our study, only one of our patients (randomized to NO-IVIG) had less than 3 of the 4 Genoa criteria.(1) Further follow-up is needed to determine if this more intensive treatment approach improves the long-term neurologic outcome of patients with neuroblastoma-associated OMA. Long-term follow up to assess neuro-developmental and learning problems of this study population is important, and there is broad agreement that the establishment of a registry would be extremely valuable. Unfortunately, the funding required to build this resource has not yet been identified. Long-term follow-up of this cohort and studies to better define the epidemiology and pathogenesis of OMA, including analysis of antineuronal antibodies in serum and cerebrospinal fluid and evaluation of magnetic resonance images of the cerebellum are secondary objectives of this study and data collection is still in progress. The results of these objectives will be the subject of separate reports. Supplementary Material This work was supported by the Chair’s Grant U10 CA-98543 and CA-180886 and Statistical and Data Center Grant U10 CA-98413 and CA-180899 of the Children’s Oncology Group from the National Cancer Institute, National Institutes of Health, Bethesda, MD.

Further follow-up is needed to determine if this more intensive treatment approach improves the long-term neurologic outcome of patients with neuroblastoma-associated OMA. Long-term follow up to assess neuro-developmental and learning problems of this study population is important, and there is broad agreement that the establishment of a registry would be extremely valuable. Unfortunately, the funding required to build this resource has not yet been identified. Long-term follow-up of this cohort and studies to better define the epidemiology and pathogenesis of OMA, including analysis of antineuronal antibodies in serum and cerebrospinal fluid and evaluation of magnetic resonance images of the cerebellum are secondary objectives of this study and data collection is still in progress. The results of these objectives will be the subject of separate reports. Supplementary Material This work was supported by the Chair’s Grant U10 CA-98543 and CA-180886 and Statistical and Data Center Grant U10 CA-98413 and CA-180899 of the Children’s Oncology Group from the National Cancer Institute, National Institutes of Health, Bethesda, MD. The authors would like to thank Dina Willis and Patrick McGrady for their assistance with data collection and analyses in the COG Statistics and Data Center at the University of Florida in Gainesville, Florida. The authors would like to dedicate this publication to the memory of Dr. Jessica A. Panzer, a coauthor and basic science investigator for OMA, who unfortunately died between the time of writing this manuscript and publication. Declaration of Interests:

The authors would like to thank Dina Willis and Patrick McGrady for their assistance with data collection and analyses in the COG Statistics and Data Center at the University of Florida in Gainesville, Florida. The authors would like to dedicate this publication to the memory of Dr. Jessica A. Panzer, a coauthor and basic science investigator for OMA, who unfortunately died between the time of writing this manuscript and publication. Declaration of Interests: All the authors equally contributed to the preparation of this manuscript. The authors have no conflicts of interest to declear. This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final citable form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain. Figure 1 Study Design Schema for COG Study ANBL00P3 * OMA evaluation time points include: at Dx, monthly for 6 months, then every other month until 1 year, 18 months, 2 years and yearly thereafter two months, six months and one year. Figure 2 CONSORT Diagram for COG Study ANBL00P3 Figure 3 a. Neuroblastoma event-free, OMA progression-free, and overall survival of the overall patient cohort (n=53) b. OMA progression-free survival for IVIG+ (n=26) versus NO-IVIG (n=27)

lities were considered when making the determination of the OMA ratings: Able to hold head consistently erect when trunk vertical? Able to reach and grasp objects with each hand? Able to roll back to front and front to back? Able to finger-feed self? Table 2 Characteristics of 53 patients with OMA on COG study ANBL00P3 IVIG + risk-based chemotherapy (IVIG+) (n=26) Risk-based chemotherapy(NO- IVIG)(n=27) Overall (n=53) Median Age at NB diagnosis [IQR] (years) 1.6 [1.3, 2.2] 1.4 [1.4, 1.9] 1.6 [1.4, 2.0] Sex Male 8 12 20 Female 18 15 33 Race White 19 17 36 African American 5 6 11 Asian 1 0 1 Unknown or Other 1 4 5 Ethnicity Hispanic 5 7 12 Non-Hispanic 20 20 40 Unknown 1 0 1 Neuroblastoma risk factors Low risk 23 21 44 Stage 1 16 16 32 Stage 2A 2 1 3 Stage 2B 5 4 9 Intermediate risk 3 4 7 Stage 2A 1 1 2 Stage 2B 1 0 1 Stage 3 1 3 4 Stage 4 0 0 0 High risk 0 2 2 Stage 3 0 1 1 Stage 4 0 1 1 Histology Favorable 16 18 34 Unfavorable 8 8 16 Unknown 2 1 3 MYCN status Non-Amplified 24 24 48 Amplified 0 2 2 Unknown 2 1 3 Age at NB diagnosis <18 months 10 15 25 ≥18 months 16 12 28

Low risk 23 21 44 Stage 1 16 16 32 Stage 2A 2 1 3 Stage 2B 5 4 9 Intermediate risk 3 4 7 Stage 2A 1 1 2 Stage 2B 1 0 1 Stage 3 1 3 4 Stage 4 0 0 0 High risk 0 2 2 Stage 3 0 1 1 Stage 4 0 1 1 Histology Favorable 16 18 34 Unfavorable 8 8 16 Unknown 2 1 3 MYCN status Non-Amplified 24 24 48 Amplified 0 2 2 Unknown 2 1 3 Age at NB diagnosis <18 months 10 15 25 ≥18 months 16 12 28 Ploidy Hyperdiploid 18 18 36 Diploid 6 7 16 Unknown 2 2 4 Stage: The patients enrolled on this study were all staged using the International Neuroblastoma Staging System (INSS), which is a surgical pathological staging system. All patients with INSS stage 1 disease have undergone complete resection of their primary unilateral localized tumor by definition. Patients with stage 2 disease have some residual tumor tissue (up to 10%) following surgical resection, while those with stages 3 and 4 disease generally undergo a biopsy only at diagnosis and have residual disease. Thus, the extent of surgical resection is captured by INSS staging. Table 3 Overall OMA response rate, by randomized treatment arm

Ploidy Hyperdiploid 18 18 36 Diploid 6 7 16 Unknown 2 2 4 Stage: The patients enrolled on this study were all staged using the International Neuroblastoma Staging System (INSS), which is a surgical pathological staging system. All patients with INSS stage 1 disease have undergone complete resection of their primary unilateral localized tumor by definition. Patients with stage 2 disease have some residual tumor tissue (up to 10%) following surgical resection, while those with stages 3 and 4 disease generally undergo a biopsy only at diagnosis and have residual disease. Thus, the extent of surgical resection is captured by INSS staging. Table 3 Overall OMA response rate, by randomized treatment arm Randomized to IVIG+ (n=26) Randomized to NO-IVIG (n=27) Total (n=53) P value OMA Responders 21 (81%) 11 (41%) 32 (60%) 0.0029* OMA Non-responders 5 (19%) 16 (59%) 21 (40%) Reasons for OMA non- response: Crossed over from NO- IVIG to IVIG+ only N/A 9 (33%) 9 (17%) Crossed over from NO- IVIG to IVIG+ then switched to ACTH N/A 3 (11%) 3 (6%) Switched to ACTH only 3 (12%) 0 (0%) 3 (6%) Stable disease 2 (8%) 4 (15%) 6 (11%) OMA Events 12 (46%) 19 (70%) 31 (58%) * In a secondary analysis of 52 patients who received therapy and completed at least one follow-up assessment of OMA response, p=0.0044. N/A-not applicable Table 4 Episodes of Grade 3, 4, or 5 Toxicity (CTCAE; version 4.0) IVIG+ (N=38)* Number of episodes (%) NO-IVIG (N=25) Number of episodes (%)

Randomized to IVIG+ (n=26) Randomized to NO-IVIG (n=27) Total (n=53) P value OMA Responders 21 (81%) 11 (41%) 32 (60%) 0.0029* OMA Non-responders 5 (19%) 16 (59%) 21 (40%) Reasons for OMA non- response: Crossed over from NO- IVIG to IVIG+ only N/A 9 (33%) 9 (17%) Crossed over from NO- IVIG to IVIG+ then switched to ACTH N/A 3 (11%) 3 (6%) Switched to ACTH only 3 (12%) 0 (0%) 3 (6%) Stable disease 2 (8%) 4 (15%) 6 (11%) OMA Events 12 (46%) 19 (70%) 31 (58%) * In a secondary analysis of 52 patients who received therapy and completed at least one follow-up assessment of OMA response, p=0.0044. N/A-not applicable Table 4 Episodes of Grade 3, 4, or 5 Toxicity (CTCAE; version 4.0) IVIG+ (N=38)* Number of episodes (%) NO-IVIG (N=25) Number of episodes (%) Hematopoietic Anemia 4 (10·5%) 3 (12·0%) Leukopenia 1 (2·6%) 1 (4·0%) Neutropenia 5 (13·2%) 4 (16·0%) Thrombocytopenia 2 (5·3%) 2 (8·0%) Febrile neutropenia 1 (2·6%) 1 (4·0%) Other 1 (2·6%) 1 (4·0%) Gastrointestinal ALT increased 1 (2·6%) AST increased 1 (2·6%) Intra-abdominal hemorrhage 1 (4·0%) Mucositis 1 (4·0%) Nausea 1 (4·0%) Vomiting 4 (10·5%) 1 (4·0%) Colitis 1 (4·0%) Diarrhea 1 (2·6%) Infectious diseses Bladder infection 2 (5·3%) 1 (4·0%) Fever 1 (2·6%) Catheter related 3 (7·9%) 2 (8·0%) Enterocolitis 1 (2·6%) 1 (4·0%) Lung infection 1 (4·0%) Other 6 (15·8%) 5 (20·0%) Nervous system Irritability 1 (2·6%) 1 (4·0%) Nystagmus 2 (5·3%) 3 (12·0%) Ataxia 1 (2·6%) 3 (12·0%) Agitation 7 (18·4%) 7 (28·0%) Personality changes 1 (2·6%) Cardiovascular Hypertension 2 (5·3%) 2 (8·0%)

Infectious diseses Bladder infection 2 (5·3%) 1 (4·0%) Fever 1 (2·6%) Catheter related 3 (7·9%) 2 (8·0%) Enterocolitis 1 (2·6%) 1 (4·0%) Lung infection 1 (4·0%) Other 6 (15·8%) 5 (20·0%) Nervous system Irritability 1 (2·6%) 1 (4·0%) Nystagmus 2 (5·3%) 3 (12·0%) Ataxia 1 (2·6%) 3 (12·0%) Agitation 7 (18·4%) 7 (28·0%) Personality changes 1 (2·6%) Cardiovascular Hypertension 2 (5·3%) 2 (8·0%) Metabolism and Nutrition Anorexia 1 (2·6%) Weight gain 1 (2·6%) 1 (4·0%) Dehydration 1 (2·6%) 1 (4·0%) Hyperglycemia 2 (5·3%) 2 (8·0%) Hypoalbuminemia 1 (2·6%) Hypoglycemia 1 (2·6%) Hypokalemia 2 (5·3%) 2 (8·0%) Hyponatremia 1 (4·0%) Hypophosphatemia 1 (2·6%) 2 (8·0%) Respiratory Hypoxemia 1 (4·0%) * The IVIG+ arm also contains patients that crossed over from the NO-IVIG arm Research in Context Evidence before this study We searched the Pubmed database using the terms opsoclonus, myoclonus, ataxia, and neuroblastoma to review prior treatment of patients with neuroblastoma and OMA. There was no date restriction used to search the database, and we did not restrict the language. Our search revealed case reports, small series, and retropsecitve studes that described neurologic response with immunosuppressive therapy in patients with neuroblastoma and OMA. These publications provide compelling evidence that OMA associated with neuroblastoma is an immune-mediated paraneoplastic syndrome. No previous prospective randomized clinical trial for this population has been reported.

scribed neurologic response with immunosuppressive therapy in patients with neuroblastoma and OMA. These publications provide compelling evidence that OMA associated with neuroblastoma is an immune-mediated paraneoplastic syndrome. No previous prospective randomized clinical trial for this population has been reported. Added value of this study This is the first and, to date, only randomized clinical trial for patients with OMA and neuroblastoma ever conducted. The results of this study demonstrate treatment effect for both arms of therapy, although significantly superior OMA response was seen with the IVIG+ regimen. As the only randomized trial with the largest number of OMA patients ever reported, the results define, for the first time, OMA response to two treatment regimens. Limitations of the study Because of the rarity of OMA, our trial has limited number of subjects and we had to accept an 80% power in our analysis. The results of our study need to be interpreted with caution in light of the small numbers of subjects but also taking into account the rarity of the disorder. In our trial, we could not evaluate the role of chemotherapy since both arms of the study contain chemotherapy with the majority of the subjects receiving cyclophosphamide.

of our study need to be interpreted with caution in light of the small numbers of subjects but also taking into account the rarity of the disorder. In our trial, we could not evaluate the role of chemotherapy since both arms of the study contain chemotherapy with the majority of the subjects receiving cyclophosphamide. Implications of all available evidence The OMS response observed with the IVIG+ regimen can be used as a historical control to investigate if treatment strategies that include new immunosuppressant agents will lead to improved outcome. Currently there is a single arm European study that is evaluating the use of rituximab in combination with dexamethasone and cyclophosphamide in the therapy of OMA.

creasing order of Socio-demographic Index in 2017. NNM=National Nutrition Mission. *In 2022 and 2030 if trends up to 2017 continue. †The state of Jammu and Kashmir was divided into two union territories in August, 2019; because we are reporting findings up to 2017, we report findings for the state of Jammu and Kashmir. Child stunting The prevalence of child stunting was 39·3% (95% UI 38·7–40·1) in India in 2017 (figure 3). This prevalence was inversely correlated with the SDI of the states (r=–0·79, p<0·0001), and varied 2·3 times between the states (figure 3). The annualised percentage reduction in stunting prevalence was seen in India in all the three periods, with the highest reduction during 2010–17 (2·63% annualised, 95% UI 2·27–2·94; figure 4B; table). The point estimate for annualised percentage reduction was higher in the high SDI compared with the low SDI state group, with the magnitude of reduction increasing over the three periods in all the SDI groups (figure 4B; table). Stunting prevalence reduced significantly in every state of India during 2010–17 (range 1·22%–3·94% annualised), but this decrease was less than the 8·6% annualised reduction needed for the NNM 2022 target and the 4·2% reduction needed for WHO and UNICEF 2030 target. The projected prevalence of 34·6% in 2022, based on trends between 1990 and 2017, was 9·6% more than the NNM target of 25·0%, and the projected prevalence of 27·7% in 2030 was 5·1% more than the WHO and UNICEF target of 22·6% (figure 5; appendix pp 42–43). The projected prevalence of stunting was higher than the target prevalence for most states of India, except for Tamil Nadu, Sikkim, Kerala, and Goa in 2022 and Uttarakhand, West Bengal, and Kerala in 2030 (figure 5; appendix pp 42–43).

Introduction Malnutrition is a major contributor to disease burden, with more than half of global deaths in children younger than 5 years of age attributable to undernutrition, the vast majority of which are in low-income and middle-income countries, including India.1, 2, 3, 4, 5 However, overweight among children is also increasing globally, including in Africa and Asia.3, 6 Addressing the challenge of malnutrition in children and women is essential to ensure optimal cognitive growth and development and overall health and productivity.7 Addressing the global burden of malnutrition is a major priority.8 To spur action and monitor progress, WHO Global Nutrition Targets were established for six malnutrition indicators to be achieved by 2025.9, 10 The UN Sustainable Development Goals (SDGs) also set targets with the aim of eliminating malnutrition by 2030.11 To strengthen the joint efforts towards reducing malnutrition worldwide, 2016–25 was declared, by the UN, as the Decade of Action on Nutrition.12 A WHO and UNICEF review in 2018 suggested that the SDG goal of eliminating all forms of malnutrition by 2030 was aspirational but not achievable and, on the basis of trends so far, recommended targets for the malnutrition indicators up to 2030.13 Research in context Evidence before this study

Addressing the global burden of malnutrition is a major priority.8 To spur action and monitor progress, WHO Global Nutrition Targets were established for six malnutrition indicators to be achieved by 2025.9, 10 The UN Sustainable Development Goals (SDGs) also set targets with the aim of eliminating malnutrition by 2030.11 To strengthen the joint efforts towards reducing malnutrition worldwide, 2016–25 was declared, by the UN, as the Decade of Action on Nutrition.12 A WHO and UNICEF review in 2018 suggested that the SDG goal of eliminating all forms of malnutrition by 2030 was aspirational but not achievable and, on the basis of trends so far, recommended targets for the malnutrition indicators up to 2030.13 Research in context Evidence before this study Existing evidence suggests that India, with a population of 1·4 billion people residing across states at varying levels of health transition, has a large and persistent burden of malnutrition, especially among children and women of reproductive age. We searched PubMed for published literature on malnutrition in India, Google for reports in the public domain, and references in these papers and reports, using the search terms “anaemia”, “breastfeeding”, “burden”, “child growth failure”, “child obesity”, “child overweight”, “DALY”, “death”, “epidemiology”, “global nutrition targets”, “India”, “infant”, “low birthweight”, “malnutrition”, “morbidity”, “mortality”, “national nutrition mission”, “neonate”, “prevalence”, “stunting”, “sustainable development goals”, “under-five”, “undernutrition”, “underweight”, and “wasting” on April 4, 2019, without language or publication date restrictions. We found several previous studies that have estimated subnational variations in malnutrition burden in India and its association with health outcomes, mainly using single data sources. However, a comprehensive understanding of the variations between the states of India in the prevalence of each malnutrition indicator, the associated deaths and disease burden, and its progress towards achieving the Indian and the global nutrition targets, using all available data sources in a single framework has not been compiled to inform relevant policy interventions suitable for the situation in each state.

valence of each malnutrition indicator, the associated deaths and disease burden, and its progress towards achieving the Indian and the global nutrition targets, using all available data sources in a single framework has not been compiled to inform relevant policy interventions suitable for the situation in each state. Added value of this study

valence of each malnutrition indicator, the associated deaths and disease burden, and its progress towards achieving the Indian and the global nutrition targets, using all available data sources in a single framework has not been compiled to inform relevant policy interventions suitable for the situation in each state. Added value of this study This study provides a comprehensive account of the burden of child and maternal malnutrition in every state of India from 1990 to 2017, by use of all available and accessible data that were analysed in the unified Global Burden of Diseases, Injuries, and Risk Factors Study framework. The findings highlight that, even with the many efforts to reduce malnutrition in India, it remains the predominant risk factor for deaths and disease burden in children younger than 5 years and the leading risk factor for disease burden in all ages combined. This study compares the projected prevalence of the malnutrition indicators in each state based on the trends up to 2017, with the targets set by the India National Nutrition Mission for 2022 and WHO and UNICEF for 2030. The substantial gaps between the trends and targets estimated in this report for most states of India indicate that progress toward all malnutrition indicators needs to be accelerated. These gaps vary between the states, indicating the extent of additional effort needed to control malnutrition in each state. The findings highlight that the modest rate of improvement in low birthweight, which is the biggest contributor among the malnutrition indicators to deaths and disease burden in children younger than 5 years of age, should be addressed through focused policy action. Besides the substantial continuing burden of poor nutrition in India, this study also reports that child overweight is increasing rapidly across all states of India.

among the malnutrition indicators to deaths and disease burden in children younger than 5 years of age, should be addressed through focused policy action. Besides the substantial continuing burden of poor nutrition in India, this study also reports that child overweight is increasing rapidly across all states of India. Implications of all available evidence Malnutrition remains one of the most serious public health challenges across India, although substantial heterogeneity exists between the states for the various malnutrition indicators and their trends over time. The resurgence in policy interest in India to reduce malnutrition across the country through the National Nutrition Mission is encouraging. This momentum can benefit from the use of state-level trends in this study, which highlight the extent of effort needed in each state to achieve the national and the global targets for the various malnutrition indicators. Decades of policy and programmatic efforts have been made in India to tackle the continuing challenge of malnutrition. In 2017, India released the National Nutrition Strategy, which outlined measures to address malnutrition across the life cycle.14 In early 2018, the Prime Minister of India launched the National Nutrition Mission (NNM), also known as POSHAN Abhiyaan, to bring focus and momentum to this effort, which has the overarching goal of reducing child and maternal malnutrition.15, 16

which outlined measures to address malnutrition across the life cycle.14 In early 2018, the Prime Minister of India launched the National Nutrition Mission (NNM), also known as POSHAN Abhiyaan, to bring focus and momentum to this effort, which has the overarching goal of reducing child and maternal malnutrition.15, 16 India had a population of 1·38 billion in 2017, spread across 29 states and seven union territories, which are at varying levels of development, leading to a heterogeneous distribution of health risks and their effects.17 The India State-Level Disease Burden Initiative has reported a varied epidemiological transition across the states of India since 1990 as part of the Global Burden of Diseases, Injuries, and Risk Factors Study (GBD).17, 18 Some subnational studies in India have reported the trends in one or more malnutrition indicators,19, 20, 21, 22, 23 and some from other countries have reported trends in malnutrition burden24, 25, 26, 27, 28 or trends in child growth failure indicators.29, 30 However, there has been no comprehensive consolidation of the malnutrition burden and the trends in all major malnutrition indicators in all states of any country using all available data sources that also relates the projected subnational trends with the policy targets for 2022 and 2030. In this report, we present consolidated findings for each state in India from 1990 to 2017 and compare these trends with Indian and global targets up to 2030 to inform state-specific policy action.

g all available data sources that also relates the projected subnational trends with the policy targets for 2022 and 2030. In this report, we present consolidated findings for each state in India from 1990 to 2017 and compare these trends with Indian and global targets up to 2030 to inform state-specific policy action. Methods Overview The analysis and findings of child and maternal malnutrition reported in this Article were produced by the India State-Level Disease Burden Initiative as part of GBD 2017. The work of this Initiative has been approved by the Health Ministry Screening Committee at the Indian Council of Medical Research and the ethics committee of the Public Health Foundation of India. A comprehensive description of the metrics, data sources, and statistical modelling for GBD 2017 has been reported elsewhere.5, 17, 18 The GBD 2017 methods relevant for this paper are summarised here and described in detail in the appendix (pp 3–26).

ch and the ethics committee of the Public Health Foundation of India. A comprehensive description of the metrics, data sources, and statistical modelling for GBD 2017 has been reported elsewhere.5, 17, 18 The GBD 2017 methods relevant for this paper are summarised here and described in detail in the appendix (pp 3–26). Estimation of exposure to malnutrition The GBD comparative risk assessment framework was used to estimate malnutrition exposure and attributable disease burden. The components of child and maternal malnutrition in GBD are described in the appendix (p 5). All accessible data sources from India were used, including national household surveys, a variety of dietary and nutrition surveys, and other epidemiological studies (appendix pp 25–37). The modelling approaches integrated multiple data inputs, using Spatiotemporal Gaussian process regression, and borrowed information across age, time, and location to produce the best possible estimates of risk exposure by location, age, sex, and year.

ion surveys, and other epidemiological studies (appendix pp 25–37). The modelling approaches integrated multiple data inputs, using Spatiotemporal Gaussian process regression, and borrowed information across age, time, and location to produce the best possible estimates of risk exposure by location, age, sex, and year. For the purpose of reporting the prevalence of the eight malnutrition indicators included in the India NNM target 2022 and the WHO and UNICEF target 2030,13, 31 the following definitions were used: low birthweight as less than 2500 g; stunting, wasting, and underweight in children younger than 5 years as height-for-age, weight-for-height, and weight-for-age below two SDs of the median in the WHO 2006 standard curve; anaemia in children younger than 5 years as haemoglobin less than 110 g/L; anaemia in women 15–49 years of age as haemoglobin less than 110 g/L if pregnant and 120 g/L if not pregnant; exclusive breastfeeding as no oral food or fluid intake during the first 6 months of life except breast milk and oral rehydration solution drops or syrups containing vitamins, minerals or medicines;32 and overweight in children aged 2–4 years as body-mass index above the monthly cutoff for normal weight as reported in the International Obesity Task Force tables.5, 33

intake during the first 6 months of life except breast milk and oral rehydration solution drops or syrups containing vitamins, minerals or medicines;32 and overweight in children aged 2–4 years as body-mass index above the monthly cutoff for normal weight as reported in the International Obesity Task Force tables.5, 33 Estimation of deaths and DALYs attributable to malnutrition Estimation of attributable disease burden included ascertainment of relative risk of disease outcomes for risk exposure-disease outcome pairs with sufficient evidence of a causal relationship in randomised controlled trials, prospective cohort studies, or case-control studies, as assessed with an approach similar to the World Cancer Research Fund grading system.5 Population attributable fractions were estimated from risk exposure, relative risks of outcomes due to exposures, and the theoretical minimum risk exposure (lowest level of risk exposure, below which its relation with a disease outcome is not supported by available evidence) for each malnutrition indicator as explained in the appendix (pp 3–24). Population attributable fractions were used to produce estimates of deaths and disability-adjusted life-years (DALYs) attributable to each malnutrition risk factor by location, age, sex, and year. DALYs are the summary measure of years of healthy life lost due to disability (YLDs) and years of life lost due to premature mortality (YLLs). The major data inputs included vital registration, verbal autopsy studies, large population-level surveys, surveillance data, and hospital-based and community-based studies (appendix pp 25–37).

mmary measure of years of healthy life lost due to disability (YLDs) and years of life lost due to premature mortality (YLLs). The major data inputs included vital registration, verbal autopsy studies, large population-level surveys, surveillance data, and hospital-based and community-based studies (appendix pp 25–37). GBD uses covariates, which are explanatory variables that have a known association with the outcome of interest, to arrive at the best possible estimate when data for the outcome are scarce but data for covariates are available.5, 34 This approach was part of the estimation process for the findings reported.

mmary measure of years of healthy life lost due to disability (YLDs) and years of life lost due to premature mortality (YLLs). The major data inputs included vital registration, verbal autopsy studies, large population-level surveys, surveillance data, and hospital-based and community-based studies (appendix pp 25–37). GBD uses covariates, which are explanatory variables that have a known association with the outcome of interest, to arrive at the best possible estimate when data for the outcome are scarce but data for covariates are available.5, 34 This approach was part of the estimation process for the findings reported. Analysis presented in this paper We report findings for 31 geographical units in India: 29 states, Union Territory of Delhi, and the union territories other than Delhi (combining the six smaller union territories of Andaman and Nicobar Islands, Chandigarh, Dadra and Nagar Haveli, Daman and Diu, Lakshadweep, and Puducherry). The state of Jammu and Kashmir was divided into two union territories in August, 2019. Because we are reporting findings up to 2017, we report findings for the state of Jammu and Kashmir. We also present findings for three groups of states categorised on the basis of their Socio-demographic Index (SDI) as calculated by GBD.35 SDI is a composite indicator of development status, which ranges from 0 to 1, and is a geometric mean of the values of the indices of lag-distributed per capita income, mean education for those 15 years of age or older, and total fertility rate in people younger than 25 years. We assessed the relationship of each malnutrition indicator with the SDI value of the states in 2017. The states were categorised into the three state groups on the basis of their SDI in 2017: low SDI (≤0·53), middle SDI (0·54–0·60), and high SDI (>0·60; appendix p 38).36

fertility rate in people younger than 25 years. We assessed the relationship of each malnutrition indicator with the SDI value of the states in 2017. The states were categorised into the three state groups on the basis of their SDI in 2017: low SDI (≤0·53), middle SDI (0·54–0·60), and high SDI (>0·60; appendix p 38).36 We assess the rates and proportion of deaths and DALYs attributable to child and maternal malnutrition among children younger than 5 years and DALYs attributable to child and maternal malnutrition among all ages in every state of India in 2017, and compare them with other risk factor categories. We also report cause-specific DALYs in children younger than 5 years attributable to malnutrition and its components in India in 2017. We present the prevalence of the eight malnutrition indicators included in Indian and global targets in the states of India. The targets set by the NNM 2022 and the WHO and UNICEF 2030 are summarised in the panel. We applied these targets to each state of India.Panel Targets set by the National Nutrition Mission for 2022 and WHO and UNICEF for 2030 National Nutrition Mission 2022 targets15, 16 • Low birthweight: 2 percentage point reduction in prevalence annually from 2017 to 2022 • Child stunting*: prevalence of 25% in 2022 • Child underweight*: 2 percentage point reduction in prevalence annually from 2017 to 2022 • Anaemia†: 3 percentage point reduction in prevalence annually in children younger than 5 years and in women 15–49 years of age from 2017 to 2022 WHO and UNICEF 2030 targets13

• Low birthweight: 2 percentage point reduction in prevalence annually from 2017 to 2022 • Child stunting*: prevalence of 25% in 2022 • Child underweight*: 2 percentage point reduction in prevalence annually from 2017 to 2022 • Anaemia†: 3 percentage point reduction in prevalence annually in children younger than 5 years and in women 15–49 years of age from 2017 to 2022 WHO and UNICEF 2030 targets13 • Low birthweight: 30% reduction in prevalence from 2012 to 2030 • Child stunting‡: 50% reduction in number of children younger than 5 years of age who are stunted from 2012 to 2030 • Child wasting: prevalence of less than 3% by 2030 • Anaemia: 50% reduction in prevalence in women 15–49 years of age from 2012 to 2030 • Breastfeeding: prevalence of exclusive breastfeeding in the first 6 months of at least 70% by 2030 • Child overweight: prevalence of less than 3% by 2030 We estimated the annualised percentage change in mid-year estimates of the prevalence of malnutrition indicators for the state SDI groups for three periods: 1990–2000, 2000–10, and 2010–17, and compared the annualised percentage change during 2010–17 with the annualised reduction needed to meet the NNM 2022 and the WHO and UNICEF 2030 targets in each state of India.

mid-year estimates of the prevalence of malnutrition indicators for the state SDI groups for three periods: 1990–2000, 2000–10, and 2010–17, and compared the annualised percentage change during 2010–17 with the annualised reduction needed to meet the NNM 2022 and the WHO and UNICEF 2030 targets in each state of India. We projected the prevalence of malnutrition indicators for India and each state up to 2030 on the basis of the trends from 1990 to 2017. The annualised change for the projections for 2018–30 was calculated using a weight function that gave higher weight to the more recent trends in each state. The detailed methods used for these projections, including the out-of-sample predictive validity test, are described in the appendix (p 23) and elsewhere.37 We report estimates with 95% uncertainty intervals (UIs) where relevant. The UIs were based on 1000 runs of the models for each quantity of interest, which have been found to be adequate for the GBD models (appendix p 23 and pp 44–49).5 The mean of these distributions was regarded as the point estimate, and the 2·5th and 97·5th percentiles were considered the 95% UI. Role of the funding source Some staff of the Indian Council of Medical Research are co-authors on this paper, having contributed to various aspects of the study and analysis. The other funder of the study had no role in the study design, data collection, data analysis, data interpretation, or writing of this paper. The corresponding author had full access to all of the data in the study and had final responsibility for the decision to submit for publication.

pects of the study and analysis. The other funder of the study had no role in the study design, data collection, data analysis, data interpretation, or writing of this paper. The corresponding author had full access to all of the data in the study and had final responsibility for the decision to submit for publication. Results Malnutrition burden Of the 1·04 million under-5 deaths in India in 2017, 706 000 (95% UI 659 000–759 000; 68·2%, 65·8–70·7) could be attributed to malnutrition.38 Although all-cause under-5 death rate in India decreased from 2336 per 100 000 (2271–2405) in 1990 to 801 per 100 000 (759–850) in 2017, the proportion of under-5 deaths attributable to malnutrition changed only modestly from 70·4% (67·0–74·0) in 1990 to 68·2% (65·8–70·7) in 2017.38 Similarly, the DALY rate attributable to malnutrition in children younger than 5 years reduced by 65·8% (62·9–68·7) from 147 956 per 100 000 (139 350–156 327) in 1990 to 50 627 (47 301–54 199) in 2017, but the proportion of total DALYs in children younger than 5 years attributable to malnutrition changed only slightly from 70·1% (66·8–70·6) in 1990 to 67·1% (64·9–69·4) in 2017, making it the predominant risk factor for health loss (appendix p 39). The vast majority of the malnutrition DALYs in children younger than 5 years in 2017 were due to mortality (94·5% of YLLs, 5·5% of YLDs).38 Although the relative contribution of child and maternal malnutrition to total DALYs across all ages has declined in India from 36·5% (95% UI 34·5–38·4) in 1990 to 17·3% (16·3–18·2) in 2017, it is still the leading risk factor for health loss (appendix p 39). The population of 1·38 billion in India in 2017 made up 18·1% of the global population, but India had 25·4% of the total global DALYs attributable to child and maternal malnutrition in 2017.38

% (95% UI 34·5–38·4) in 1990 to 17·3% (16·3–18·2) in 2017, it is still the leading risk factor for health loss (appendix p 39). The population of 1·38 billion in India in 2017 made up 18·1% of the global population, but India had 25·4% of the total global DALYs attributable to child and maternal malnutrition in 2017.38 Malnutrition was the leading risk factor in children younger than 5 years in every state of India in 2017 (appendix p 39). The DALY rate attributable to malnutrition in children younger than 5 years varied 6·8 times between the states, and it was 1·8 times higher in the low SDI than in the middle SDI state groups and 2·4 times higher than in high SDI state groups (figure 1, appendix p 39). Malnutrition was also the leading risk factor across all ages in 23 states that comprised 64% of India's population in 2017, contributing 10·0%–26·4% of the total DALYs (appendix p 40). The DALY rate attributable to malnutrition across all ages varied 6·0 times between states, and it was 2·0 times higher in the low SDI than in the middle SDI state groups and 2·7 times higher than in high SDI state groups (appendix p 40).Figure 1 Disability-adjusted life-years rate attributable to malnutrition in children younger than 5 years of age in the states of India, 2017 The state of Jammu and Kashmir was divided into two union territories in August 2019; because we are reporting findings up to 2017, we report findings for the state of Jammu and Kashmir.

Malnutrition was the leading risk factor in children younger than 5 years in every state of India in 2017 (appendix p 39). The DALY rate attributable to malnutrition in children younger than 5 years varied 6·8 times between the states, and it was 1·8 times higher in the low SDI than in the middle SDI state groups and 2·4 times higher than in high SDI state groups (figure 1, appendix p 39). Malnutrition was also the leading risk factor across all ages in 23 states that comprised 64% of India's population in 2017, contributing 10·0%–26·4% of the total DALYs (appendix p 40). The DALY rate attributable to malnutrition across all ages varied 6·0 times between states, and it was 2·0 times higher in the low SDI than in the middle SDI state groups and 2·7 times higher than in high SDI state groups (appendix p 40).Figure 1 Disability-adjusted life-years rate attributable to malnutrition in children younger than 5 years of age in the states of India, 2017 The state of Jammu and Kashmir was divided into two union territories in August 2019; because we are reporting findings up to 2017, we report findings for the state of Jammu and Kashmir. The highest proportion of the malnutrition DALYs in children younger than 5 years in India in 2017 was from low birthweight and short gestation (43·6%, 95% UI 41·8–45·2) followed by child growth failure (20·7%, 19·0–22·5; appendix p 41). Of the total DALYs attributable to malnutrition in children younger than 5 years in India in 2017, the largest proportions were from neonatal disorders (54·9%) followed by lower respiratory infections (22·6%) and diarrhoeal diseases (13·3%; figure 2). The highest proportion of DALYs attributable to low birthweight and short gestation were from neonatal disorders (84·7%; figure 2). The highest proportion of DALYs attributable to child growth failure were from lower respiratory infections (47·0%) followed by diarrhoeal diseases (35·3%; figure 2). The DALYs attributable to suboptimal breastfeeding were from diarrhoeal diseases (62·1%) and lower respiratory infections (37·9%; figure 2).Figure 2 Cause-specific disability-adjusted life-years attributable to malnutrition in children younger than 5 years of age in India, 2017

by diarrhoeal diseases (35·3%; figure 2). The DALYs attributable to suboptimal breastfeeding were from diarrhoeal diseases (62·1%) and lower respiratory infections (37·9%; figure 2).Figure 2 Cause-specific disability-adjusted life-years attributable to malnutrition in children younger than 5 years of age in India, 2017 Data are presented for child and maternal malnutrition and the three leading components. Data shown are percent of total disability-adjusted life-years for each risk that manifests through different diseases. Protein-energy malnutrition is a specific disease cause in Global Burden of Diseases, Injuries, and Risk Factors Study, as opposed to the malnutrition risk factor indicators. For child and maternal malnutrition, the other category includes childhood infections other than diarrhoeal diseases and lower respiratory infections, vitamin A deficiency, and sudden infant death syndrome. For child growth failure, the other category includes measles.

pposed to the malnutrition risk factor indicators. For child and maternal malnutrition, the other category includes childhood infections other than diarrhoeal diseases and lower respiratory infections, vitamin A deficiency, and sudden infant death syndrome. For child growth failure, the other category includes measles. Low birthweight The prevalence of low birthweight in India was 21·4% (95% UI 20·8–21·9) in 2017. This prevalence decreased moderately with increasing SDI of states (r=–0·38, p=0·034), and varied 2·8 times between the states (figure 3). Low birthweight prevalence decreased modestly in India in all the three periods, with relatively higher decline during 2010–17 (1·12% annualised, 95% UI 0·68–1·57; figure 4A; table). The point estimate of annualised percentage reduction was highest in the high SDI state group, with the magnitude of reduction increasing over the three periods across the SDI groups (figure 4A; table). Low birthweight prevalence decreased significantly in 14 states of India during 2010–17 (range 1·10%–3·76% annualised) but was much lower than the 11·8% annualised reduction needed for the NNM 2022 target (table). None of the states except Sikkim had the annualised reduction of 2·3% needed for the WHO and UNICEF 2030 target. The projected prevalence, based on trends between 1990 and 2017, of 20·3% in 2022 was 2·9% more than the NNM target of 11·4%, and the projected prevalence of 18·7% in 2030 was 11·4% more than the WHO and UNICEF target of 15·8% (figure 5; appendix pp 42–43). The projected prevalence of low birthweight was higher than the target prevalence in 2022 for all states and in 2030 for all states except Sikkim and Maharashtra (figure 5; appendix pp 42–43).Figure 3 Prevalence of malnutrition indicators in the states of India, 2017

NICEF target of 15·8% (figure 5; appendix pp 42–43). The projected prevalence of low birthweight was higher than the target prevalence in 2022 for all states and in 2030 for all states except Sikkim and Maharashtra (figure 5; appendix pp 42–43).Figure 3 Prevalence of malnutrition indicators in the states of India, 2017 The states are listed in increasing order of Socio-demographic Index in 2017. The population of each state SDI group in 2017 is shown in parentheses. UI=uncertainty interval. SDI=Socio-demographic Index. *The state of Jammu and Kashmir was divided into two union territories in August, 2019; because we are reporting findings up to 2017, we report findings for the state of Jammu and Kashmir. Figure 4 Annualised percentage change in mid-year estimates of the prevalence of malnutrition indicators in the states of India grouped by SDI, 1990–2000, 2000–10, and 2010–17 (A) Low birthweight. (B) Child stunting. (C) Child wasting. (D) Child underweight. (E) Child anaemia. (F) Anaemia in women. (G) Exclusive breastfeeding. (H) Child overweight. Error bars represent 95% uncertainty intervals. SDI=Socio-demographic Index. Table Annualised percentage change in the prevalence of malnutrition indicators in the states of India, 2010–17

(A) Low birthweight. (B) Child stunting. (C) Child wasting. (D) Child underweight. (E) Child anaemia. (F) Anaemia in women. (G) Exclusive breastfeeding. (H) Child overweight. Error bars represent 95% uncertainty intervals. SDI=Socio-demographic Index. Table Annualised percentage change in the prevalence of malnutrition indicators in the states of India, 2010–17 Low birthweight (95% UI) Child stunting (95% UI) Child wasting (95% UI) Child underweight (95% UI) Child anaemia (95% UI) Anaemia in women (95% UI) Exclusive breastfeeding (95% UI) Child overweight (95% UI) India −1·12% (−1·57 to −0·68) −2·63% (−2·94 to −2·27) −1·23% (−1·47 to −0·97) −3·22% (−3·44 to −2·98) −1·81% (−2·26 to −1·36) −0·68% (−0·89 to −0·44) 1·19% (0·22 to 2·16) 4·98% (2·18 to 7·78) Low SDI −1·03% (−1·73 to −0·48) −2·34% (−2·83 to −1·81) −0·44% (−0·85 to −0·07) −3·41% (−3·77 to −3·04) −1·66% (−2·35 to −1·00) −0·98% (−1·35 to −0·60) 1·13% (0·08 to 2·21) 5·43% (2·48 to 8·39) Bihar −1·27% (−2·62 to 0·13) −1·82% (−2·97 to −0·38) −1·69% (−2·76 to −0·68) −3·42% (−4·36 to −2·54) −1·07% (−2·84 to 0·84) −0·36% (−1·22 to 0·45) 2·58% (0·99 to 4·38) 5·04% (0·94 to 9·03) Madhya Pradesh −0·94% (−2·10 to 0·28) −2·93% (−3·70 to −2·10) −1·71% (−2·53 to −0·94) −4·78% (−5·76 to −3·88) −1·83% (−3·20 to −0·53) −1·75% (−2·51 to −0·99) 2·66% (1·16 to 4·35) 7·21% (3·35 to 11·24) Jharkhand −1·28% (−2·29 to −0·34) −1·63% (−2·59 to −0·56) −0·30% (−1·13 to 0·55) −3·15% (−4·20 to −2·19) −0·88% (−2·33 to 0·73) −0·46% (−1·21 to 0·29) 0·87% (−0·13 to 1·88) 5·98% (1·73 to 9·99) Uttar Pradesh −0·77% (−2·21 to 0·77) −2·02% (−3·02 to −0·88) 0·55% (−0·37 to 1·44) −3·03% (−3·76 to −2·36) −0·88% (−2·17 to 0·43) −0·53% (−1·35 to 0·38) −0·47% (−2·04 to 1·17) 5·08% (1·31 to 9·01) Rajasthan −1·05% (−2·05 to 0·12) −3·03% (−3·69 to −2·33) 0·84% (−0·02 to 1·71) −3·31% (−3·86 to −2·77) −1·37% (−3·02 to 0·13) −1·44% (−2·33 to −0·52) 2·22% (0·73 to 3·95) 4·36% (0·39 to 8·25) Chhattisgarh −2·09% (−3·82 to −0·56) −3·51% (−4·63 to −2·27) 0·23% (−0·73 to 1·24) −3·48% (−4·29 to −2·79) −4·43% (−6·50 to −2·28) −1·91% (−2·86 to −0·90) 0·28% (−0·23 to 0·84) 6·32% (2·44 to 10·22) Odisha −1·15% (−2·18 to 0·04) −3·15% (−3·97 to −2·23) −1·14% (−1·98 to −0·26) −2·94% (−3·68 to −2·32) −5·03% (−7·09 to −3·02) −0·98% (−2·15 to 0·13) 1·40% (0·32 to 2·56) 6·03% (1·84 to 9·94) Assam −1·27% (−2·50 to −0·10) −2·73% (−3·80 to −1·60) −1·06% (−2·24 to 0·08) −2·61% (−3·20 to −2·08) −7·27% (−9·55 to −4·84) −2·78% (−3·99 to −1·76) 0·87% (−0·15 to 2·00) 4·94% (1·01 to 9·05) Middle SDI −1·28% (−1·75 to −0·88) −3·18% (−3·51 to

to −3·02) −0·98% (−2·15 to 0·13) 1·40% (0·32 to 2·56) 6·03% (1·84 to 9·94) Assam −1·27% (−2·50 to −0·10) −2·73% (−3·80 to −1·60) −1·06% (−2·24 to 0·08) −2·61% (−3·20 to −2·08) −7·27% (−9·55 to −4·84) −2·78% (−3·99 to −1·76) 0·87% (−0·15 to 2·00) 4·94% (1·01 to 9·05) Middle SDI −1·28% (−1·75 to −0·88) −3·18% (−3·51 to −2·82) −2·16% (−2·51 to −1·81) −2·93% (−3·15 to −2·72) −1·99% (−2·73 to −1·23) −0·61% (−0·97 to −0·22) 0·94% (0·09 to 1·85) 5·58% (2·73 to 8·46) Andhra Pradesh −1·29% (−2·30 to 0·03) −2·87% (−3·66 to −2·12) −1·34% (−2·20 to −0·44) −2·69% (−3·20 to −2·21) −1·79% (−3·71 to 0·20) −0·17% (−1·02 to 0·59) 0·52% (−0·27 to 1·40) 5·54% (1·61 to 9·57) West Bengal −1·45% (−2·72 to −0·39) −3·92% (−4·81 to −3·07) −3·51% (−4·37 to −2·56) −2·67% (−3·18 to −2·16) −1·66% (−3·90 to 0·52) −0·33% (−1·29 to 0·73) 0·60% (−0·81 to 1·99) 5·89% (2·18 to 9·55) Tripura −1·37% (−2·44 to −0·46) −2·89% (−3·81 to −1·94) −1·94% (−2·98 to −0·92) −2·73% (−3·48 to −2·01) −3·06% (−5·55 to −0·84) −1·20% (−2·39 to 0·08) 2·75% (1·12 to 4·46) 4·29% (0·04 to 7·92) Arunachal Pradesh −0·64% (−1·68 to 0·70) −2·98% (−3·87 to −2·02) −0·75% (−1·76 to 0·29) −1·96% (−2·66 to −1·34) −1·53% (−3·19 to 0·22) −1·92% (−3·00 to −0·82) 1·06% (−0·29 to 2·52) 4·06% (0·36 to 7·66) Meghalaya −1·23% (−2·54 to 0·28) −1·22% (−1·97 to −0·44) −4·04% (−5·25 to −2·87) −5·37% (−6·32 to −4·45) −4·04% (−7·12 to −1·54) −0·23% (−1·16 to 0·78) 2·88% (−0·15 to 6·28) 2·63% (−1·16 to 6·50) Karnataka −1·27% (−2·39 to −0·43) −2·94% (−3·65 to −2·27) −2·09% (−2·95 to −1·23) −3·17% (−3·61 to −2·75) −2·18% (−3·88 to −0·54) −0·85% (−1·77 to 0·06) 0·44% (−0·57 to 1·52) 5·53% (1·70 to 9·40) Telangana −0·99% (−3·06 to 1·13) −3·63% (−4·67 to −2·64) −3·07% (−4·05 to −2·17) −4·20% (−4·75 to −3·66) −2·17% (−3·95 to −0·43) −0·91% (−1·89 to 0·06) 1·07% (0·19 to 2·09) 5·38% (1·60 to 9·23) Gujarat −1·32% (−2·27 to −0·34) −2·86% (−3·74 to −2·01) −0·47% (−1·15 to 0·23) −2·45% (−3·03 to −1·92) −3·16% (−4·50 to −1·87) −1·09% (−1·82 to −0·32) 1·07% (−0·28 to 2·54) 5·74% (2·03 to 9·57) Manipur −1·84% (−3·49 to −0·82) −2·29% (−3·13 to −1·50) −1·93% (−3·35 to −0·60) −2·88% (−3·55 to −2·20) −5·52% (−8·87 to −2·46) −2·74% (−4·29 to −1·05) 0·96% (0·22 to 1·76) 5·66% (1·56 to 9·65) Jammu and Kashmir* −1·36% (−2·35 to −0·03) −2·51% (−3·37 to −1·67) −0·41% (−1·67 to 0·86) −3·18% (−3·79 to −2·55) −0·81% (−2·60 to 1·16) −0·53% (−1·44 to 0·40) 1·49% (0·34 to 2·71) 4·84% (0·99 to 8·83) Haryana −1·10% (−1·84 to −0·23) −3·42% (−4·17 to −2·61) −3·66% (−4·90 to −2·53) −3·20% (

to 1·76) 5·66% (1·56 to 9·65) Jammu and Kashmir* −1·36% (−2·35 to −0·03) −2·51% (−3·37 to −1·67) −0·41% (−1·67 to 0·86) −3·18% (−3·79 to −2·55) −0·81% (−2·60 to 1·16) −0·53% (−1·44 to 0·40) 1·49% (0·34 to 2·71) 4·84% (0·99 to 8·83) Haryana −1·10% (−1·84 to −0·23) −3·42% (−4·17 to −2·61) −3·66% (−4·90 to −2·53) −3·20% ( −3·68 to −2·70) −0·06% (−1·48 to 1·31) −0·17% (−1·07 to 0·62) 3·36% (1·25 to 5·67) 6·70% (2·47 to 10·83) High SDI −1·52% (−2·16 to −0·74) −3·33% (−3·72 to −2·92) −2·01% (−2·46 to −1·58) −3·11% (−3·36 to −2·84) −2·16% (−2·98 to −1·33) −0·21% (−0·60 to 0·25) 1·75% (0·54 to 3·01) 4·83% (1·88 to 7·90) Uttarakhand −0·64% (−1·65 to 0·17) −3·87% (−4·67 to −3·08) −1·60% (−2·84 to −0·38) −4·83% (−5·39 to −4·31) −0·02% (−1·80 to 1·75) −1·40% (−2·40 to −0·48) 2·15% (0·33 to 4·06) 7·12% (3·16 to 11·07) Tamil Nadu −1·72% (−3·57 to 0·16) −3·31% (−4·12 to −2·42) −1·97% (−2·71 to −1·23) −3·47% (−3·93 to −3·01) −2·74% (−4·52 to −0·94) 0·32% (−0·45 to 1·08) 2·04% (0·39 to 3·86) 4·96% (1·06 to 9·05) Mizoram −0·87% (−4·09 to 1·74) −2·88% (−3·79 to −2·04) −1·18% (−2·18 to −0·11) −2·38% (−3·01 to −1·75) −8·35% (−11·25 to −5·29) −3·00% (−4·44 to −1·57) 1·91% (0·38 to 3·67) 2·49% (−1·45 to 6·44) Maharashtra −2·14% (−3·04 to −1·07) −3·26% (−4·03 to −2·50) −2·67% (−3·44 to −1·92) −2·84% (−3·28 to −2·39) −2·47% (−4·12 to −0·88) −0·69% (−1·54 to 0·29) 1·44% (0·08 to 2·99) 4·17% (0·42 to 8·18) Punjab −1·44% (−2·47 to −0·45) −3·37% (−4·13 to −2·60) −1·90% (−3·00 to −0·80) −3·52% (−3·97 to −3·07) −1·75% (−3·32 to −0·10) 0·18% (−0·68 to 1·07) 2·15% (0·40 to 4·29) 4·86% (0·76 to 8·92) Sikkim −3·76% (−7·24 to −1·29) −3·15% (−4·21 to −2·23) −2·35% (−3·88 to −0·82) −3·41% (−4·21 to −2·64) −1·79% (−4·04 to 0·28) −3·40% (−4·64 to −2·07) 2·80% (0·65 to 5·04) 2·86% (−1·08 to 7·15) Nagaland −1·17% (−2·75 to −0·15) −2·68% (−3·52 to −1·88) −1·34% (−2·48 to −0·30) −2·99% (−3·62 to −2·39) −6·17% (−9·71 to −2·37) −3·45% (−4·72 to −2·10) 2·25% (0·05 to 4·67) 2·87% (−0·95 to 6·63) Himachal Pradesh −0·99% (−2·38 to 0·26) −3·18% (−4·26 to −2·16) −1·95% (−3·32 to −0·56) −3·87% (−4·61 to −3·24) −0·72% (−2·68 to 1·46) −0·05% (−1·03 to 0·96) 3·81% (1·85 to 6·10) 6·27% (2·56 to 10·21) UTs other than Delhi −1·10% (−3·63 to 1·63) −2·48% (−3·95 to −1·01) −0·53% (−1·95 to 0·77) −1·60% (−3·14 to −0·57) 1·01% (−1·19 to 2·93) 0·42% (−0·57 to 1·54) 1·01% (−0·34 to 2·54) 4·56% (0·74 to 8·29) Kerala −0·84% (−4·11 to 1·51) −3·94% (−4·92 to −2·95) −0·27% (−1·22 to 0·66) −3·89% (−4·49 to −3·29) −3·68% (−5·93 to −1·42) −0·58% (−1

lhi −1·10% (−3·63 to 1·63) −2·48% (−3·95 to −1·01) −0·53% (−1·95 to 0·77) −1·60% (−3·14 to −0·57) 1·01% (−1·19 to 2·93) 0·42% (−0·57 to 1·54) 1·01% (−0·34 to 2·54) 4·56% (0·74 to 8·29) Kerala −0·84% (−4·11 to 1·51) −3·94% (−4·92 to −2·95) −0·27% (−1·22 to 0·66) −3·89% (−4·49 to −3·29) −3·68% (−5·93 to −1·42) −0·58% (−1 ·85 to 0·74) 0·92% (−0·38 to 2·28) 6·15% (2·45 to 10·14) Delhi −0·31% (−1·76 to 1·16) −3·29% (−4·29 to −2·37) −1·99% (−3·31 to −0·77) −1·85% (−2·50 to −1·21) 0·38% (−1·40 to 2·30) 0·31% (−0·75 to 1·36) 2·83% (0·27 to 5·48) 5·11% (1·49 to 8·87) Goa −0·88% (−2·71 to 0·33) −2·41% (−3·80 to −1·17) 0·32% (−0·73 to 1·36) −1·87% (−2·53 to −1·22) −0·01% (−2·60 to 2·46) −1·84% (−3·13 to −0·59) 3·97% (1·51 to 6·46) 4·61% (0·63 to 8·43) The states are listed in increasing order of Socio-demographic Index in 2017. UI=uncertainty interval. SDI=Socio-demographic Index. UTs=Union Territiories. * The state of Jammu and Kashmir was divided into two union territories in August 2019; as we are reporting findings up to 2017, we report findings for the state of Jammu and Kashmir. Figure 5 Gap between projected prevalence of malnutrition indicators and the National Nutrition Mission 2022 and the WHO and UNICEF 2030 targets in the states of India The states are listed in increasing order of Socio-demographic Index in 2017. NNM=National Nutrition Mission. *In 2022 and 2030 if trends up to 2017 continue. †The state of Jammu and Kashmir was divided into two union territories in August, 2019; because we are reporting findings up to 2017, we report findings for the state of Jammu and Kashmir.

was 5·1% more than the WHO and UNICEF target of 22·6% (figure 5; appendix pp 42–43). The projected prevalence of stunting was higher than the target prevalence for most states of India, except for Tamil Nadu, Sikkim, Kerala, and Goa in 2022 and Uttarakhand, West Bengal, and Kerala in 2030 (figure 5; appendix pp 42–43). Child wasting Within child growth failure, the highest contribution to DALYs was from child wasting (19·0%, 95% UI 16·2–21·2; appendix p 40). The prevalence of child wasting was 15·7% (95% UI 15·6–15·9) in India in 2017. This prevalence did not have a significant correlation with the SDI of states (r=–0·30, p=0·097), but had a 3·1 times variation between the states (figure 3). The point estimate of annualised percentage reduction of wasting in India was highest during 2010–17 (1·23%, 95% UI 0·97–1·47), with substantial variation across the state SDI groups during the three periods (figure 4C; table). The annualised percentage decrease was similar across the state SDI groups during 1990–2000, was highest in the low SDI state group during 2000–10, and was higher in the middle and high SDI groups than the low SDI group during 2010–17 (figure 4C; table). Although wasting prevalence significantly declined in many states of India, the reduction was much lower than the 12·0% annualised reduction needed for the WHO and UNICEF 2030 target (table). The projected prevalence fro India of 13·4% in 2030, based on trends between 1990 and 2017, was 10·4% higher than the WHO and UNICEF target of wasting prevalence of less than 3% (figure 5; appendix p 43). No state met these targets.

than the 12·0% annualised reduction needed for the WHO and UNICEF 2030 target (table). The projected prevalence fro India of 13·4% in 2030, based on trends between 1990 and 2017, was 10·4% higher than the WHO and UNICEF target of wasting prevalence of less than 3% (figure 5; appendix p 43). No state met these targets. Child underweight The prevalence of child underweight was 32·7% (95% UI 32·3–33·1) in India in 2017. This prevalence was inversely correlated with the SDI of the states (r=–0·76, p<0·0001), and varied 2·6 times between the states (figure 3). The annualised percentage reduction in underweight prevalence was seen in India in all the three periods, with higher reductions occurring in the last two periods than in 1990–2000 (figure 4D; table). The point estimate for annualised percentage reduction was higher in the high SDI state group compared with the low SDI group during 1990–2000 and 2000–10 but was higher in the low SDI group compared with the high SDI group during 2010–17 (figure 4D; table). The underweight prevalence reduced significantly in every state of India during 2010–17 (range 1·60%–5·37% annualised), but this decrease was less than the 7·0% annualised reduction needed to achieve the NNM 2022 target. The projected prevalence for India of 27·5% in 2022, based on trends between 1990 and 2017, was 4·8% more than the NNM target of 22·7%; this difference varied from 2·1% to 8·5% across the states (figure 5; appendix p 42).

t this decrease was less than the 7·0% annualised reduction needed to achieve the NNM 2022 target. The projected prevalence for India of 27·5% in 2022, based on trends between 1990 and 2017, was 4·8% more than the NNM target of 22·7%; this difference varied from 2·1% to 8·5% across the states (figure 5; appendix p 42). Child anaemia The prevalence of child anaemia was 59·7% (95% UI 56·2–63·8) in India in 2017. This prevalence did not have a significant correlation with the SDI of the states (r=–0·25, p=0·17), but had a 3·5 times variation between the states (figure 3). The annualised percentage prevalence of child anaemia decreased in India during 2010–17 (1·81%, 95% UI 1·36–2·26), with no significant change during 2000–10 (figure 4E; table). The estimate of child anaemia prevalence decreased significantly in the high SDI state group during 2000–10 and decreased in all SDI groups during 2010–17. Although the prevalence of child anaemia decreased significantly in 16 states of India during 2010–17 (range 1·75%–8·35% annualised), none of these states, except Assam, Mizoram, and Nagaland, had the annualised reduction of 5·6% needed to achieve the NNM 2022 target (table). The projected prevalence of 56·4% in India in 2022, based on trends between 1990 and 2017, was 11·7% higher than the NNM target of 44·7%; this difference was more than 10% for most of the states (figure 5; appendix p 42).

and Nagaland, had the annualised reduction of 5·6% needed to achieve the NNM 2022 target (table). The projected prevalence of 56·4% in India in 2022, based on trends between 1990 and 2017, was 11·7% higher than the NNM target of 44·7%; this difference was more than 10% for most of the states (figure 5; appendix p 42). Anaemia in women The prevalence of anaemia in women 15–49 years of age was 54·4% (95% UI 53·7–55·2) in India in 2017. This prevalence was inversely correlated with the SDI of the states (r=–0·40, p=0·027), and varied 2·3 times between the states (figure 3). The annualised percentage of anaemia prevalence decreased in India during 2010–17 (0·68%, 95% UI 0·44–0·89), with no change during 2000–10 (figure 4F; table). The point estimate of anaemia prevalence decreased in all the SDI state groups in all the three periods, except for the middle SDI group during 2000–10. The annualised percentage decrease was highest in the high SDI state group during 1990–2000, and in the low and middle SDI groups during 2010–17 (figure 4F; table). The prevalence of anaemia decreased significantly in 12 states of India during 2010–17 (range 1·09%–3·45% annualised), but none of the states had the annualised reduction of 6·2% needed to achieve the NNM 2022 target and 4·9% for the WHO and UNICEF 2030 target (table). The projected prevalence of 53·2% in 2022, based on trends between 1990 and 2017, was 13·8% higher than the NNM target of 39·4%, and the projected prevalence of 51·1% in 2030 was 22·8% higher than the WHO and UNICEF target of 28·3%; these gaps varied substantially across the states of India (figure 5; appendix pp 42–43).

e projected prevalence of 53·2% in 2022, based on trends between 1990 and 2017, was 13·8% higher than the NNM target of 39·4%, and the projected prevalence of 51·1% in 2030 was 22·8% higher than the WHO and UNICEF target of 28·3%; these gaps varied substantially across the states of India (figure 5; appendix pp 42–43). Exclusive breastfeeding The prevalence of exclusive breastfeeding was 53·3% (95% UI 51·5–54·9) in India in 2017, with a moderate inverse correlation with the SDI of the states (r=–0·38, p=0·036). This prevalence varied 2·2 times between the states (figure 3). The annualised percentage increase in the prevalence of exclusive breastfeeding in India during 2010–17 (1·19%, 95% UI 0·22–2·16) was similar to 1990–2000 (1·04%, 0·26–1·82; figure 4G; table). Except for low SDI state group during 2000–10, the prevalence of exclusive breastfeeding increased in all the SDI groups in all the three periods, with relatively higher increase in the high SDI group (figure 4G; table). However, based on the modest increasing trends between 1990 and 2017, the projected prevalence for India was 59·3%, 10·7% less than the WHO and UNICEF 2030 target of at least 70%; only a few states met this target (figure 5; appendix p 43).

iods, with relatively higher increase in the high SDI group (figure 4G; table). However, based on the modest increasing trends between 1990 and 2017, the projected prevalence for India was 59·3%, 10·7% less than the WHO and UNICEF 2030 target of at least 70%; only a few states met this target (figure 5; appendix p 43). Child overweight The prevalence of overweight in children aged 2–4 years was 11·5% (95% UI 8·5–14·9) in India in 2017. This prevalence was positively correlated with the SDI of the states (r=0·79, p<0·0001), with 3·4 times variation between the states. The prevalence of child overweight increased significantly in India during 2010–17 (4·98%, 95% UI 2·18–7·78), with similar annualised percentage increase in the three state SDI groups (figure 4H; table). Significant annualised percentage increase occurred in the middle SDI and high SDI state groups during 2000–10 also. The projected child overweight prevalence of 17·5% in India in 2030, based on trends between 1990 and 2017, was 14·5% higher than the WHO and UNICEF 2030 target of less than 3% (figure 5; appendix p 43), and no state met these targets.

e increase occurred in the middle SDI and high SDI state groups during 2000–10 also. The projected child overweight prevalence of 17·5% in India in 2030, based on trends between 1990 and 2017, was 14·5% higher than the WHO and UNICEF 2030 target of less than 3% (figure 5; appendix p 43), and no state met these targets. Discussion The findings in this report provide insights into the trends in child and maternal malnutrition burden and the key indicators that can inform further efforts to reduce this burden for every state of India. Although the burden of child and maternal malnutrition has declined in India since 1990, it remains the predominant risk factor for health loss in children younger than 5 years of age in every state of the country and the leading risk factor for health loss across all ages in the majority of states. The malnutrition DALY rate is highest in the low SDI states, with substantial variation between the states. The malnutrition DALYs in children younger than 5 years of age are predominantly due to premature mortality.

e of the country and the leading risk factor for health loss across all ages in the majority of states. The malnutrition DALY rate is highest in the low SDI states, with substantial variation between the states. The malnutrition DALYs in children younger than 5 years of age are predominantly due to premature mortality. Low birthweight, the largest contributor to the malnutrition DALYs in India, had a prevalence of 21% in 2017, which showed a modest declining trend. Within child growth failure, the highest contribution to DALYs was from wasting, the prevalence of which declined only moderately in India during 2010–17. The prevalence of stunting and underweight has been decreasing, however, the prevalence has remained very high in India at 39% and 33%, respectively, in 2017. The prevalence of anaemia has been extremely high in India at 60% in children and 54% in women in 2017, with only moderate decline during 2010–17. However, the prevalence of child overweight has increased considerably in India in the past decade, with a prevalence of 12% in 2017. The prevalence of exclusive breastfeeding was 53% in India in 2017, with a moderate increase during 2010–17. Substantial state-level variations exist in the prevalence for each of the malnutrition indicators. The findings in this report indicate that, if the trends up to 2017 continue, the NNM 2022 and the WHO and UNICEF 2030 targets will not be achieved in most states of India, except for low birthweight and stunting in a few states and exclusive breastfeeding in several.

prevalence for each of the malnutrition indicators. The findings in this report indicate that, if the trends up to 2017 continue, the NNM 2022 and the WHO and UNICEF 2030 targets will not be achieved in most states of India, except for low birthweight and stunting in a few states and exclusive breastfeeding in several. Because low birthweight was the largest contributor to child malnutrition DALYs in India, its slow decline should be addressed as a priority. South Asia, with India as its largest component, is estimated to have the highest prevalence of low birthweight for any region in the world.39 A major issue with tracking low birthweight is the poor quality of birthweight data in many low-income and middle-income countries, including India.39 Low birthweight adversely affects not only child health but also increases the risk of chronic diseases later in life.40, 41, 42, 43, 44, 45, 46, 47 Weight at birth is an intergenerational issue dependent on an interplay of various factors, including maternal undernutrition, intrauterine growth, gestation at birth, birth spacing and order, and maternal age. The higher proportion of underweight women in the reproductive age group in India compared with sub-Saharan Africa has been suggested to contribute to a higher prevalence of low birthweight in India, even though sub-Saharan Africa is poorer.48 Chronic energy deficiency in women of reproductive age is a manifestation of long-standing malnutrition reported to be common in India, which increases the risk of preterm births and infants with low birthweight.1, 49, 50, 51 Improving the nutritional status of girls in general and that of women in the preconception period and during pregnancy and provision of quality antenatal care, including the treatment of pregnancy complications, would positively affect low birthweight and extend the benefits to the next generation.39, 52, 53, 54 Aligned with the Global Every Newborn Action Plan, the India Newborn Action Plan launched in 2014 aims to reduce low birthweight through improved preconception and antenatal care, adolescent-specific health services, nutritional counselling, and micronutrient supplementation.46, 55

to the next generation.39, 52, 53, 54 Aligned with the Global Every Newborn Action Plan, the India Newborn Action Plan launched in 2014 aims to reduce low birthweight through improved preconception and antenatal care, adolescent-specific health services, nutritional counselling, and micronutrient supplementation.46, 55 India has been trying to address child malnutrition for many decades through various policy initiatives, such as the Integrated Child Development Scheme launched in 1975, the National Nutrition Policy 1993, the Mid Day Meal Scheme for school children 1995, and the National Food Security Act 2013,56, 57 but the prevalence of stunting, wasting and underweight remains high. The prevalence of stunting, an indicator of chronic undernutrition, caused by a variety of social, environmental, and economic risk factors, is unsurprisingly highest in the less developed states. However, the prevalence of wasting, indicative of acute undernutrition, is highest in some of the more developed states. The decline in stunting is usually accompanied by a temporary increase or stagnancy in wasting; therefore, achieving a simultaneous reduction of stunting and wasting might be difficult.58 Women's status, birth order, son preference, and open defecation contribute to relatively higher rates of undernutrition among children in South Asian countries, including India, compared with sub-Saharan African countries with comparable or lower incomes.59, 60, 61 Alongside the nutrient-based interventions, more comprehensive and inclusive policies addressing all of the key determinants of child malnutrition are needed to accelerate reduction of child growth failure in India as also envisioned in the NNM.15

ub-Saharan African countries with comparable or lower incomes.59, 60, 61 Alongside the nutrient-based interventions, more comprehensive and inclusive policies addressing all of the key determinants of child malnutrition are needed to accelerate reduction of child growth failure in India as also envisioned in the NNM.15 The high burden of anaemia in children and women, with only a modest decline since 1990, is a major public health issue in India. Anaemia increases the risk of adverse birth outcomes and mortality during and after child birth and leads to poor cognitive and physical development and mortality in children.1, 62, 63, 64, 65, 66 Interventions to improve nutrition of girls, including reduction of the prevalence of anaemia, starting at a young age, are needed for better pregnancy-related and early child health outcomes and for a beneficial long-term effect on future generations.1, 53, 54, 67, 68 India launched the National Iron Plus Initiative in 2013 to comprehensively address anaemia burden across the life cycle, through age-specific interventions with iron and folic acid supplementation and deworming.69 Other initiatives in India to address the developmental needs of adolescents in general, and the nutrition and reproductive health needs of adolescent girls in particular, include the National Adolescent Health Programme 2014 and the Scheme for Adolescent Girls.70, 71 As emphasised in the recent NNM of India, a set of interventions to optimise the health of adolescents and young women would be more effective than any single intervention addressing macronutrient or micronutrient deficiency.15

ude the National Adolescent Health Programme 2014 and the Scheme for Adolescent Girls.70, 71 As emphasised in the recent NNM of India, a set of interventions to optimise the health of adolescents and young women would be more effective than any single intervention addressing macronutrient or micronutrient deficiency.15 Many states in India will not be able to meet the WHO and UNICEF 2030 target of 70% exclusive breastfeeding if the slow rate of increase observed up to 2017 continues. Promotion of exclusive breastfeeding is essential to support optimal growth and development of the infant and address the burden of child growth failure and child overweight.72, 73 The efforts to increase exclusive breastfeeding in India include the Infant and Young Child Feeding Guidelines, government regulation on breast milk substitutes, and operational platforms to deliver interventions, such as the Integrated Child Development Scheme and the National Breastfeeding Promotion Programme.74, 75, 76, 77, 78 Challenges in further improving the rates of exclusive breastfeeding in India include societal beliefs that encourage mixed feeding practices, inadequate lactation support, and aggressive promotion of breast milk substitutes.73

ild Development Scheme and the National Breastfeeding Promotion Programme.74, 75, 76, 77, 78 Challenges in further improving the rates of exclusive breastfeeding in India include societal beliefs that encourage mixed feeding practices, inadequate lactation support, and aggressive promotion of breast milk substitutes.73 The increasing prevalence of overweight in children in India is of concern, with adverse effects on health during childhood as well as long-term chronic effects persisting into adulthood.1 Interventions to reduce the burden of overweight children in India should focus on improving the modifiable risk factors, including appropriate child feeding practices, dietary intake, and physical activity.79 The draft of the Food Safety and Standards (Safe and Wholesome Food for School Children) Regulations 2018 is indicative of efforts to promote a balanced diet and reduce the availability of foods high in fat, salt, and sugar to school-aged children in India.80 However, a comprehensive approach to addressing child overweight needs to be developed in India.

ndards (Safe and Wholesome Food for School Children) Regulations 2018 is indicative of efforts to promote a balanced diet and reduce the availability of foods high in fat, salt, and sugar to school-aged children in India.80 However, a comprehensive approach to addressing child overweight needs to be developed in India. Substantial improvements across the malnutrition indicators in the states of India would require an integrated nutrition policy to effectively address the broader determinants of undernutrition across the life cycle. These improvements include providing clean drinking water, reducing rates of open defecation, improving women's status, enhancing agricultural productivity and food security, promoting nutrition-sensitive agriculture, coupled with harmonisation of efforts across ministries and sectors, political will and good governance, and strategic investments in a multisectoral approach.1, 59, 77, 81, 82, 83, 84, 85 The Government of India launched a revamped NNM with a budget of US$1·3 billion to comprehensively address the challenge of persistent undernutrition.15, 16, 86 The goal of this Mission is to systematically synergise a variety of nutrition-related activities of various government ministries and stakeholders in order to strengthen many maternal and child health initiatives across the life cycle. This includes the supplementary nutrition component of Integrated Child Development Scheme, Maternity Benefit Programme, Mid Day Meal Scheme, dietary diversification to improve iron and folic acid intake, engaging the private sector in food fortification efforts, and placing emphasis on the broader social determinants of nutrition. This renewed focus on a multisectoral approach to address malnutrition is encouraging, and the targets set by the Mission could motivate the states to accelerate progress. Additionally, several ongoing initiatives under the Ministry of Women and Child Development to reduce gender inequality and empower women can also contribute to improvements in malnutrition.87 The major ongoing sanitation improvement drive in India under the Swachh Bharat Mission is also expected to contribute to the reduction in malnutrition. However, our findings suggest that the malnutrition indicator targets set by NNM for 2022 are aspirational, and the rate of improvement needed to achieve these targets is much higher than the rate observed in this study, which might be difficult to reach in a short period.

o contribute to the reduction in malnutrition. However, our findings suggest that the malnutrition indicator targets set by NNM for 2022 are aspirational, and the rate of improvement needed to achieve these targets is much higher than the rate observed in this study, which might be difficult to reach in a short period. This slow pace of improvement needs to be accelerated, so that future prevalence of the malnutrition indicators are better than our projections based on trends so far. Just as the WHO and UNICEF 2030 targets were set with the realisation that the SDG target of eliminating all forms of malnutrition by 2030 was not achievable, by use of the trends presented for each state in this report, the NNM could set bold but potentially achievable targets for 2030 for India. Another report has also estimated a substantial gap between the NNM 2022 target for stunting and the projected prevalence for India if the trends between National Family Health Survey 2005–06 and 2015–16 continue.23 Generally, low-income and middle-income countries would benefit from setting national and subnational targets for reducing malnutrition that are based on robust analysis of trends.

get for stunting and the projected prevalence for India if the trends between National Family Health Survey 2005–06 and 2015–16 continue.23 Generally, low-income and middle-income countries would benefit from setting national and subnational targets for reducing malnutrition that are based on robust analysis of trends. The general limitations of the GBD methods have been described elsewhere.5 Limitations specific to the findings in this report include relatively poor data on low birthweight in India. Birthweight is generally not recorded well or remembered by parents and is incorrectly documented in many instances in India, so relatively less reliable data on this indicator is obtained in household surveys; this situation needs to improve for better estimates of low birthweight.39, 88 There is also scope for improving the estimates of preterm births in India with more robust data. GBD defines child overweight at age 2–4 years using the International Obesity Task Force standards as more data from various countries are available for these ages, which generally leads to relatively higher child overweight prevalence as compared with the WHO child growth reference data for children younger than 5 years of age used by the WHO global nutrition target.89 We used periods to present the rate of change in trends for malnutrition indicators, because they are easy to understand. However, this approach could mask finer trends within these periods. The strengths of the findings in this report include the use of all accessible data sources in India and modelling them for best fits, which reduces the chance of erratic estimates that can be observed in individual surveys with variable data quality.90, 91, 92 The estimates of malnutrition burden and the trends in its indicators for every state of the country over a quarter of a century and their future trajectory produced using the standardised GBD methods, and the comprehensive inputs by leading experts in India on the analysis and interpretation of the findings, are other strengths of the findings presented in this report.

in its indicators for every state of the country over a quarter of a century and their future trajectory produced using the standardised GBD methods, and the comprehensive inputs by leading experts in India on the analysis and interpretation of the findings, are other strengths of the findings presented in this report. India has had increasing food self-sufficiency and food security with the Green Revolution that started in the late 1960s.81, 93, 94 Even with these improvements, India continues to have a high prevalence of undernutrition, combined with an increasing prevalence of overweight and obesity in a subset of the population. Addressing this persistent development problem requires India to ensure implementation of practical and effective policies and interventions across the life cycle that consider the subnational variations and the context of each state. The focus brought to malnutrition by the National Nutrition Mission effort is likely to build momentum towards more rapid reduction of malnutrition in India. The findings in this report provide a reference for monitoring the progress of malnutrition indicators in the coming years in each state of the country. Robust estimation of malnutrition indicators and their trends over time would also be needed at the district level to understand intra-state variations, especially in the large states of India. Comprehensive subnational assessment of the trends in malnutrition indicators, their projections, and their association with policy targets, as presented in this report, could also be useful in other countries to inform decision making to improve subnational disparities in nutritional status.

the large states of India. Comprehensive subnational assessment of the trends in malnutrition indicators, their projections, and their association with policy targets, as presented in this report, could also be useful in other countries to inform decision making to improve subnational disparities in nutritional status. This online publication has been corrected. The corrected version first appeared at thelancet.com/child-adolescent on September 30, 2019 Supplementary Material Supplementary appendix

the large states of India. Comprehensive subnational assessment of the trends in malnutrition indicators, their projections, and their association with policy targets, as presented in this report, could also be useful in other countries to inform decision making to improve subnational disparities in nutritional status. This online publication has been corrected. The corrected version first appeared at thelancet.com/child-adolescent on September 30, 2019 Supplementary Material Supplementary appendix Acknowledgments The research reported in this Article was funded by the Bill & Melinda Gates Foundation and the Indian Council of Medical Research, Department of Health Research, Government of India. SS was employed by the Indian Council of Medical Research during the initial phase of this work and is now employed by WHO. The content of this publication is solely the responsibility of the authors and does not necessarily represent the official views of the Gates Foundation, the Government of India, or WHO. We gratefully acknowledge the Ministry of Health and Family Welfare of the Government of India for its support and encouragement of the India State-Level Disease Burden Initiative, the governments of the states of India for their support of this work, the many institutions and investigators across India who provided data and other inputs for this study, the valuable guidance of the Advisory Board of this Initiative, and the large number of staff at the Indian Council of Medical Research, Public Health Foundation of India, and the Institute for Health Metrics and Evaluation for their contribution to various aspects of the work of this Initiative.

inputs for this study, the valuable guidance of the Advisory Board of this Initiative, and the large number of staff at the Indian Council of Medical Research, Public Health Foundation of India, and the Institute for Health Metrics and Evaluation for their contribution to various aspects of the work of this Initiative. India State-Level Disease Burden Initiative Malnutrition Collaborators Soumya Swaminathan, Rajkumar Hemalatha, Anamika Pandey, Nicholas J Kassebaum, Avula Laxmaiah, Thingnganing Longvah, Rakesh Lodha, Siddarth Ramji, G Anil Kumar, Ashkan Afshin, Subodh S Gupta, Anita Kar, Ajay K Khera, Matthews Mathai, Shally Awasthi, Reeta Rasaily, Chris M Varghese, Anoushka I Millear, Helena Manguerra, William M Gardner, Reed Sorenson, Mari J Sankar, Manorama Purwar, Melissa Furtado, *Priyanka G Bansal, *Ryan Barber, *Joy K Chakma, *Julian Chalek, *Supriya Dwivedi, *Nancy Fullman, *Brahmam N Ginnela, *Scott D Glenn, *William Godwin, *Zaozianlungliu Gonmei, *Rachita Gupta, *Suparna G Jerath, *Rajni Kant, *Varsha Krish, *Rachakulla H Kumar, *Laishram Ladusingh, *Indrapal I Meshram, *Parul Mutreja, *Balakrishna Nagalla, *Arlappa Nimmathota, *Christopher M Odell, *Helen E Olsen, *Ashalata Pati, *Brandon Pickering, *Kankipati V Radhakrishna, *Neena Raina, *Zane Rankin, *Deepika Saraf, *R S Sharma, *Anju Sinha, *Bhaskar Varanasi, Chander Shekhar, Hendrik J Bekedam, K Srinath Reddy, Stephen S Lim, Simon I Hay, Rakhi Dandona, Christopher J L Murray, G S Toteja, Lalit Dandona. *Names listed alphabetically.

India State-Level Disease Burden Initiative Malnutrition Collaborators Soumya Swaminathan, Rajkumar Hemalatha, Anamika Pandey, Nicholas J Kassebaum, Avula Laxmaiah, Thingnganing Longvah, Rakesh Lodha, Siddarth Ramji, G Anil Kumar, Ashkan Afshin, Subodh S Gupta, Anita Kar, Ajay K Khera, Matthews Mathai, Shally Awasthi, Reeta Rasaily, Chris M Varghese, Anoushka I Millear, Helena Manguerra, William M Gardner, Reed Sorenson, Mari J Sankar, Manorama Purwar, Melissa Furtado, *Priyanka G Bansal, *Ryan Barber, *Joy K Chakma, *Julian Chalek, *Supriya Dwivedi, *Nancy Fullman, *Brahmam N Ginnela, *Scott D Glenn, *William Godwin, *Zaozianlungliu Gonmei, *Rachita Gupta, *Suparna G Jerath, *Rajni Kant, *Varsha Krish, *Rachakulla H Kumar, *Laishram Ladusingh, *Indrapal I Meshram, *Parul Mutreja, *Balakrishna Nagalla, *Arlappa Nimmathota, *Christopher M Odell, *Helen E Olsen, *Ashalata Pati, *Brandon Pickering, *Kankipati V Radhakrishna, *Neena Raina, *Zane Rankin, *Deepika Saraf, *R S Sharma, *Anju Sinha, *Bhaskar Varanasi, Chander Shekhar, Hendrik J Bekedam, K Srinath Reddy, Stephen S Lim, Simon I Hay, Rakhi Dandona, Christopher J L Murray, G S Toteja, Lalit Dandona. *Names listed alphabetically. Affiliations Indian Council of Medical Research, New Delhi, India (S Swaminathan MD, R Rasaily PhD, P G Bansal PhD, J K Chakma MD, S Dwivedi PhD, Z Gonmei PhD, R Kant PhD, D Saraf PhD, R S Sharma PhD, A Sinha PhD, C Shekhar MD, G S Toteja PhD); World Health Organization, Geneva, Switzerland (S Swaminathan); National Institute of Nutrition, Indian Council of Medical Research, Hyderabad, India (R Hemalatha MD, A Laxmaiah PhD, T Longvah PhD, B N Ginnela DPH, R H Kumar DPH, I I Meshram MD, B Nagalla PhD, A Nimmathota MD, K V Radhakrishna DCH, B Varanasi MSc); Public Health Foundation of India, Gurugram, India (A Pandey PhD, G A Kumar PhD, C M Varghese MPH, M Furtado MPH, P Mutreja MA, Prof K S Reddy DM, Prof R Dandona PhD, Prof L Dandona MD); Institute for Health Metrics and Evaluation, University of Washington, Seattle, WA, USA (N J Kassebaum MD, A Afshin MD, A I Millear MPH, H Manguerra BS, W M Gardner AB, R Sorenson MPH, R Barber BS, J Chalek BS, N Fullman MPH, S D Glenn MSc, W Godwin MPH, V Krish BA, C M Odell MPP, H E Olsen MA, B Pickering BS, Z Rankin MPH, Prof S S Lim PhD, Prof S I Hay FMedSci, Prof R Dandona, Prof C J L Murray MD, Prof L Dandona); Department of Anesthesiology and Pain Medicine, University of Washington, Seattle, WA, USA (N J Kassebaum); Department of Paediatrics, All India Institute of Medical Sciences, New Delhi, India (Prof R Lodha MD, M J Sankar DM); Department of Paediatrics, Maulana Azad Medical College, New Delhi, India (Prof S Ramji MD); Department of Community Medicine, Mahatma Gandhi Institute of Medical Sciences, Wardha, India (Prof S S Gupta MD); School of Health Sciences, Savitribai Phule Pune University, Pune, India (Prof A Kar PhD); Ministry of Health and Family Welfare, Government of India, New Delhi, India (A K Khera MD, A Pati MPH); Centre for Maternal and Newborn Health, Liverpool School of Tropical Medicine, Liverpool, UK (Prof M Mathai PhD); Department of Pediatrics, King George's Medical University, Lucknow, India (Prof S Awasthi MD); Nagpur INTERGROWTH-21st Research Centre, Ketkar Hospital, Nagpur, India (M Purwar MD); Indian Institute of Public Health—Delhi, Public Health Foundation of India, Gurugram, India (S G Jerat

Liverpool, UK (Prof M Mathai PhD); Department of Pediatrics, King George's Medical University, Lucknow, India (Prof S Awasthi MD); Nagpur INTERGROWTH-21st Research Centre, Ketkar Hospital, Nagpur, India (M Purwar MD); Indian Institute of Public Health—Delhi, Public Health Foundation of India, Gurugram, India (S G Jerat h, PhD); WHO India Country Office, New Delhi, India (R Gupta PhD, H J Bekedam MD); Bodoland University, Kokrajhar, India (Prof L Ladusingh PhD); Regional Office for South-East Asia, World Health Organization, New Delhi, India (N Raina PhD). Contributors LD and SS conceptualised this paper and drafted it with contributions from RH, AP, NJK, AL, TL, GAK, CMV, MF, and RD. The other authors provided data, participated in the analysis, or reviewed the findings (or a combination of these) and contributed to the interpretation. All authors agreed with the final version of the paper. Declaration of interests SS, RH, AL, TL, RR, PGB, JKC, SD, BNG, ZG, RK, RHK, IIM, BN, AN, KVR, DS, RSS, AS, BV, CS, and GST are or have been employees of the Indian Council of Medical Research, which partially funded this research. All other authors declare no competing interests. * The National Nutrition Mission 2022 target for stunting and underweight is for children aged 0–6 years; for consistency with the global targets we estimated this for children younger than 5 years. † The National Nutrition Mission 2022 target for child anaemia is for children aged 6–59 months; for consistency with the other targets we estimated this for children younger than 5 years.

* The National Nutrition Mission 2022 target for stunting and underweight is for children aged 0–6 years; for consistency with the global targets we estimated this for children younger than 5 years. † The National Nutrition Mission 2022 target for child anaemia is for children aged 6–59 months; for consistency with the other targets we estimated this for children younger than 5 years. ‡ We estimated a relative reduction in the prevalence of stunting instead of the absolute numbers for consistency with other indicators, because all other targets are based on prevalence.

Introduction An estimated 2·5 million neonatal deaths occurred in 2018, accounting for 47% of deaths among children younger than 5 years.1 The burden of neonatal mortality is unequally distributed, with nearly 80% of these deaths occurring in sub-Saharan Africa and southern Asia.1 Between 2000 and 2015, neonatal mortality declined more slowly than mortality among children aged 1–59 months. This disparity was particularly notable in sub-Saharan Africa, where the annual mortality reduction for newborn babies was less than half of that for 1–59 month-olds.2 Slow progress in this region might be related to high incidences of preterm birth (<37 completed weeks of gestation) and low birthweight (≤2000 g),3, 4 poor access to care for neonates,5, 6 and health system capacity issues, including shortages of skilled providers, essential supplies, and basic equipment.6, 7, 8 Research in context Evidence before this study

Introduction An estimated 2·5 million neonatal deaths occurred in 2018, accounting for 47% of deaths among children younger than 5 years.1 The burden of neonatal mortality is unequally distributed, with nearly 80% of these deaths occurring in sub-Saharan Africa and southern Asia.1 Between 2000 and 2015, neonatal mortality declined more slowly than mortality among children aged 1–59 months. This disparity was particularly notable in sub-Saharan Africa, where the annual mortality reduction for newborn babies was less than half of that for 1–59 month-olds.2 Slow progress in this region might be related to high incidences of preterm birth (<37 completed weeks of gestation) and low birthweight (≤2000 g),3, 4 poor access to care for neonates,5, 6 and health system capacity issues, including shortages of skilled providers, essential supplies, and basic equipment.6, 7, 8 Research in context Evidence before this study 2·5 million neonatal deaths occur each year, among which 78% are in sub-Saharan Africa and southern Asia. More than 80% of these deaths occur in babies with a low birthweight who are small because they are preterm, small for their gestational age, or both. There has been slow progress in reducing neonatal mortality, which accounts for nearly half of deaths in children younger than 5 years, highlighting the need for scale-up of effective interventions for neonates who are at risk. We searched PubMed for studies published between Jan 1, 1992, and July 31, 2019, with the search terms “infant, newborn”, and “infant mortality”, or “infant, newborn, diseases, and mortality”, or “infant, premature, diseases, and mortality”, or “hospital mortality”, and “severity of illness index”, or “risk assessment”, or “predictive value of tests”, or “outcome assessment”. Multiple risk scores for neonatal mortality, illness severity, and clinical instability have been developed for intensive care settings. Most of these risk scores are not feasible for low-income and middle-income countries, because they rely on laboratory-derived and therapy-derived parameters that are frequently unavailable, or on clinical observations that are not reliably measurable. A need remains for a highly predictive tool, feasible for use in resource-constrained settings, to help providers objectively assess mortality risk in the most vulnerable babies.

oratory-derived and therapy-derived parameters that are frequently unavailable, or on clinical observations that are not reliably measurable. A need remains for a highly predictive tool, feasible for use in resource-constrained settings, to help providers objectively assess mortality risk in the most vulnerable babies. Added value of this study

oratory-derived and therapy-derived parameters that are frequently unavailable, or on clinical observations that are not reliably measurable. A need remains for a highly predictive tool, feasible for use in resource-constrained settings, to help providers objectively assess mortality risk in the most vulnerable babies. Added value of this study Using population-wide data from 110 176 neonates admitted to 187 hospitals across England and Wales, this study has derived and validated a mortality risk score for neonates weighing 2000 g or less (NMR-2000). To our knowledge, this is the largest dataset used to develop and validate a neonatal mortality risk score. NMR-2000 uses data on three parameters: birthweight, oxygen saturation (peripheral capillary oxygen) at admission, and highest level of respiratory support at any point within 24 h of birth. The model had very good discrimination and goodness-of-fit across the development and UK validation samples, with a c-index of 0·8859–0·8930 and a Brier score of 0·0232–0·0271. The simplified integer score, which can be measured and calculated at the bedside, showed predictive ability similar to the model using regression coefficients. In the Gambian dataset, which included 550 neonates at one hospital, the model had good discrimination and overall goodness-of-fit, with a c-index of 0·8170 and a Brier score of 0·1688. The simplified integer score showed similar performance, with a c-index of 0·8082. Complete data for scoring were available for 83% of neonates. These findings indicate that the NMR-2000 is valid for use in health facilities where pulse oximetry is available and underscore the fact that implementation in low-income and middle-income countries would require sensitisation regarding documentation of the three parameters used in the model.

able for 83% of neonates. These findings indicate that the NMR-2000 is valid for use in health facilities where pulse oximetry is available and underscore the fact that implementation in low-income and middle-income countries would require sensitisation regarding documentation of the three parameters used in the model. Implications of all the available evidence To reduce neonatal mortality worldwide, there is an urgent need to scale-up evidence-based interventions targeting the major causes of death. Our risk score could expedite recognition of severe illness and enable targeted delivery of care to small and vulnerable neonates, increasing effectiveness and efficiency of facility-based neonatal care in low-income and middle-income countries. Further research is required to validate NMR-2000 in low-resource settings using a larger sample, and to evaluate its usefulness for clinical decision making. The score has the potential to inform resource use, including nursing workload.

ciency of facility-based neonatal care in low-income and middle-income countries. Further research is required to validate NMR-2000 in low-resource settings using a larger sample, and to evaluate its usefulness for clinical decision making. The score has the potential to inform resource use, including nursing workload. More than 80% of neonatal deaths in sub-Saharan Africa and southern Asia occur in babies with a low birthweight.9 Low birthweight can result from being preterm, being small for their gestational age, or both. Mortality is twice as high in full-term neonates who are small for their gestational age than in full-term neonates who are of average size, and 15 times higher in preterm neonates who are small for their gestational age than in babies with either characteristic alone.10 The lower the birthweight and gestational age, the higher the mortality risk.9 Around 86% of neonates born at fewer than 28 weeks' gestation, and 41% of those born at 28–31 weeks, will die without access to intensive care;11 more than 75% of neonates in sub-Saharan Africa and southern Asia have no access to such care.7 Estimates suggest that neonatal special care,5, 7 including resuscitation, kangaroo mother care, feeding support or intravenous fluids, and management of respiratory distress, infections, and jaundice, could prevent 70% of preterm deaths and decrease prematurity-related causes of neonatal mortality by 58%.12

uch care.7 Estimates suggest that neonatal special care,5, 7 including resuscitation, kangaroo mother care, feeding support or intravenous fluids, and management of respiratory distress, infections, and jaundice, could prevent 70% of preterm deaths and decrease prematurity-related causes of neonatal mortality by 58%.12 Various systems for scoring illness severity and mortality risk in neonates have been developed, primarily for high-income settings (appendix p 6). Therapy-based approaches, such as the Neonatal Therapeutic Intervention Scoring System (NTISS),13 categorise illness severity by the quantity and type of therapies administered. By contrast, the Score for Neonatal Acute Physiology (SNAP),14, 15 the Transport Risk Index of Physiologic Stability (TRIPS),16 and other physiology-based approaches use objective, measurable parameters that vary with illness severity, such as blood pressure. Related models, such as the Clinical Risk Index for Babies (CRIB),17, 18 combine physiological parameters with perinatal factors, such as birthweight, to provide an overall mortality risk score. SNAP and CRIB are the most widely used systems and have been extensively validated.19

lness severity, such as blood pressure. Related models, such as the Clinical Risk Index for Babies (CRIB),17, 18 combine physiological parameters with perinatal factors, such as birthweight, to provide an overall mortality risk score. SNAP and CRIB are the most widely used systems and have been extensively validated.19 Notably, none of the aforementioned systems are practicable for routine use in low-income and middle-income countries (LMICs), because these systems rely on laboratory-derived and therapy-derived measures that are often not available, or on clinical observations that are not reliably measurable in these settings.20, 21, 22 The simplified age-weight-sex (SAWS) score is the only validated neonatal mortality score designed for low-resource settings. Among a derivation cohort of 428 neonates weighing 1500 g or less in Bangladesh and Egypt, the SAWS was reported to have moderate discrimination for in-hospital mortality.22 To improve the quality of facility-based neonatal care in LMICs, a highly predictive tool, which is feasible for routine use, is needed to help providers objectively assess mortality risk in small babies.

or less in Bangladesh and Egypt, the SAWS was reported to have moderate discrimination for in-hospital mortality.22 To improve the quality of facility-based neonatal care in LMICs, a highly predictive tool, which is feasible for routine use, is needed to help providers objectively assess mortality risk in small babies. This study has two parts: (1) model development using data from the UK, and (2) model validation using data from the UK and The Gambia. The objectives were to evaluate existing neonatal illness severity and mortality risk scores to select candidate variables for use in the new model; develop and validate a score feasible for use in LMICs to predict in-hospital neonatal mortality risk among neonates weighing 2000 g or less within 24 h of birth; and compare the performance of the novel score (NMR-2000) with that of an existing score (CRIB-II).

s to select candidate variables for use in the new model; develop and validate a score feasible for use in LMICs to predict in-hospital neonatal mortality risk among neonates weighing 2000 g or less within 24 h of birth; and compare the performance of the novel score (NMR-2000) with that of an existing score (CRIB-II). Methods Study design and participants This retrospective study used data held in the UK National Neonatal Research Database (NNRD) from 187 neonatal units to develop a model for scoring in-hospital neonatal mortality risk in LMICs. The NNRD holds de-identified patient-level data, recorded by health-care providers as part of routine care, from admissions to National Health Service neonatal units in England starting from 2008, and in Wales and Scotland starting from 2012. This study included neonates admitted to units in England and Wales between Jan 1, 2010, and Dec 31, 2017 (appendix p 1). The following exclusion criteria were also applied: birthweight more than 2000 g; being admitted at older than 6 h or following discharge home; neonates who were stillborn; neonates who died in the delivery room; neonates who were moribund (received only comfort care before death; appendix p 1).

1, 2017 (appendix p 1). The following exclusion criteria were also applied: birthweight more than 2000 g; being admitted at older than 6 h or following discharge home; neonates who were stillborn; neonates who died in the delivery room; neonates who were moribund (received only comfort care before death; appendix p 1). As well as data from the NNRD, we used data on neonates in The Gambia. West and central Africa have the highest neonatal mortality worldwide (31 in 1000 livebirths).1 In The Gambia in 2018, the neonatal mortality (26 in 1000 livebirths) ranked ninth among the 16 countries of west Africa.1 An estimated 12% of Gambian neonates are born preterm.3 Edward Francis Small Teaching Hospital (EFSTH) in Banjul is the national referral hospital where the neonatal unit admits around 1400 neonates annually. From 2010 to 2013, case-fatality was 35% overall and prematurity-related complications were the leading cause of death.23 The Gambian cohort included all neonates weighing less than 2000 g who were admitted to EFSTH between May 1, 2018, and Sept 30, 2019, who were screened for but not enrolled in the Early KMC (eKMC) trial (NCT03555981). Some routine data, including mode of delivery and treatments administered, were collected from medical charts by trained study personnel. Other data collected as part of the screening process were exported from the trial database, including birthweight, sex, birth plurality, referral status, and peripheral capillary oxygen saturation (SpO2; appendix p 1).

delivery and treatments administered, were collected from medical charts by trained study personnel. Other data collected as part of the screening process were exported from the trial database, including birthweight, sex, birth plurality, referral status, and peripheral capillary oxygen saturation (SpO2; appendix p 1). Model development and validation using the UK dataset was approved by the North West–Preston Research Ethics Committee (17/NW/0709), the UK Health Research Authority, and the London School of Hygiene and Tropical Medicine (LSHTM; reference number 14594). Letters were sent to the UK Neonatal Collaborative Lead of all units contributing data to the NNRD, providing information about the study and giving each an opportunity to opt out. Model validation using the Gambian dataset was approved by research ethics committees of the Gambian Government and Medical Research Council Unit The Gambia at the LSHTM (reference number 1643) and LSHTM (reference number 16189). Consent was not obtained, as this was a retrospective study using de-identified data.

out. Model validation using the Gambian dataset was approved by research ethics committees of the Gambian Government and Medical Research Council Unit The Gambia at the LSHTM (reference number 1643) and LSHTM (reference number 16189). Consent was not obtained, as this was a retrospective study using de-identified data. Selection of candidate variables To select the candidate variables for the model, 18 studies describing existing systems for assessing neonatal mortality risk and illness severity were reviewed to generate a list of potential parameters (appendix p 6). Parameters that are typically unavailable, infrequently obtained, or unreliably measured in low-resource settings were excluded (appendix p 1). Remaining parameters were evaluated using the following exclusion criteria: low prevalence in the NNRD (<0·1%); high proportion of missing data in the development dataset (≥20%); not predictive of mortality in neonates who are preterm or have a low birthweight; low prevalence within the first 24 h of life; little evidence to support validity; and concept better represented by an alternative variable (appendix pp 7–9).

D (<0·1%); high proportion of missing data in the development dataset (≥20%); not predictive of mortality in neonates who are preterm or have a low birthweight; low prevalence within the first 24 h of life; little evidence to support validity; and concept better represented by an alternative variable (appendix pp 7–9). Model development To create the model, we used a development sample of neonates admitted to a random sample of neonatal units in England and Wales from Jan 1, 2010, to Dec 31, 2016. Logistic regression models were derived to model in-hospital mortality risk. Robust standard errors allowed for clustering within units. All candidate variables were included in a complete multivariable model, which was progressively simplified using reverse stepwise selection, with the least statistically significant variable removed at each step. Discrimination was assessed with the c-index, equivalent to the area under the receiver operating characteristic (ROC) curve. A value of 0·5 indicates no predictive ability, 0·8 is considered good, and 1 is perfect.21 Overall goodness-of-fit was assessed with the Brier score and calibration using plots of observed versus predicted risk (appendix p 2). Multiple imputation with chained equations was used to assess the effect of missing data (appendix p 2). The logistic regression model was executed across the imputed datasets, and the resulting β coefficients and c-index were compared with original estimates. A sensitivity analysis excluding neonates whose admission age was unknown (anonymised data derived from calculated difference between birth time and admission time) was done to reassess model performance, because admission at more than 6 h of age was an exclusion criterion. Performance was additionally reassessed following exclusion of neonates who were transferred for any reason because outcome data were not available for these babies. Performance for predicting mortality within 24 h of birth was evaluated in a secondary analysis, because 36% of neonatal deaths occur within this timeframe.24

ormance was additionally reassessed following exclusion of neonates who were transferred for any reason because outcome data were not available for these babies. Performance for predicting mortality within 24 h of birth was evaluated in a secondary analysis, because 36% of neonatal deaths occur within this timeframe.24 Score development To develop the score, we assigned the parameters in the final model points proportional to their β regression coefficient values. Whole numbers were used to generate an easily calculable score. We arbitrarily defined low-risk, medium-risk, and high-risk groups (appendix pp 2–3). To assess the calibration of the score to the model using regression coefficients, observed risks in groups and population deciles of scores were derived and compared with mean predicted risks in each group or population decile. We assessed overall predictive ability of the score using the c-index. Model validation We then evaluated both the internal and external validity of the model. Internal validity is the reproducibility of a prediction model for the underlying population from which the data originated.25 Bootstrap resampling with 1000 samples from within the development sample was used to internally validate the model, estimating optimism-adjusted measures of discrimination and goodness-of-fit in each bootstrap sample (appendix p 3). Performance of the refitted model in each bootstrap sample was compared with that of the refitted model in the original development sample; estimates of optimism were averaged and subtracted to provide optimism-adjusted measures.

ed measures of discrimination and goodness-of-fit in each bootstrap sample (appendix p 3). Performance of the refitted model in each bootstrap sample was compared with that of the refitted model in the original development sample; estimates of optimism were averaged and subtracted to provide optimism-adjusted measures. External validity is the generalisability of a model's performance to related populations.25 The model was evaluated in three external validation samples: the random sample, which included neonates admitted to the units withheld from the development sample; the temporal sample, which included neonates admitted to units in England and Wales from Jan 1, to Dec 31, 2017; and the Gambian sample, which included neonates admitted to EFSTH between May 1, 2018, and Sept 30, 2019. Each sample was used to assess distinctive features of model performance. The random sample tested performance in different care settings in the UK within the same timeframe, whereas the temporal sample tested performance during a later timeframe. The Gambian sample was used to test performance in a LMIC care setting. We assessed model performance in each validation sample separately and in the UK full (combined random and temporal samples) validation sample. Discrimination was evaluated using the c-index, and goodness-of-fit was evaluated using the Brier score. Calibration was assessed by plotting observed versus predicted risk. We assessed the overall predictive ability of the risk score using the c-index. In the Gambian sample, we redefined low-risk, medium-risk, and high-risk groups to account for increased case fatality in this sample compared with the UK samples (appendix p 3). Observed risks in groups and population deciles of scores were derived and compared with mean predicted risks in each group or population decile of the Gambian sample.

redefined low-risk, medium-risk, and high-risk groups to account for increased case fatality in this sample compared with the UK samples (appendix p 3). Observed risks in groups and population deciles of scores were derived and compared with mean predicted risks in each group or population decile of the Gambian sample. Comparison with the CRIB II score The NNRD did not include all the variables required for calculation of CRIB, SNAP, SNAP-II, SNAPPE-II, TRIPS, or TRIPS-II scores (appendix pp 7–9); therefore, CRIB-II was selected for comparison with NMR-2000 (appendix p 3). Because CRIB-II has only been validated for use in neonates born up to 32 weeks' gestation,18 we compared c-indices for CRIB-II and NMR-2000 among neonates born at 32 weeks' gestation or earlier in the full validation sample. All analyses were completed using Stata (version 15). Role of the funding source The funders of the study had no role in study design, data collection, data analysis, data interpretation, or writing of the report. The corresponding author had full access to all the data in the study and had final responsibility for the decision to submit for publication.

Comparison with the CRIB II score The NNRD did not include all the variables required for calculation of CRIB, SNAP, SNAP-II, SNAPPE-II, TRIPS, or TRIPS-II scores (appendix pp 7–9); therefore, CRIB-II was selected for comparison with NMR-2000 (appendix p 3). Because CRIB-II has only been validated for use in neonates born up to 32 weeks' gestation,18 we compared c-indices for CRIB-II and NMR-2000 among neonates born at 32 weeks' gestation or earlier in the full validation sample. All analyses were completed using Stata (version 15). Role of the funding source The funders of the study had no role in study design, data collection, data analysis, data interpretation, or writing of the report. The corresponding author had full access to all the data in the study and had final responsibility for the decision to submit for publication. Results For the selection of candidate variables, 18 studies were reviewed to generate a list of 84 potential parameters. 45 (53·6%) of 84 parameters were considered infeasible for LMICs and were excluded, among which, 25 (55·5%) also had a low prevalence or were not included in the NNRD (figure 1). Eight (9·5%) of 84 parameters were excluded because the evidence was scarce, because the parameters had poor predictive ability in preterm neonates or neonates with low birthweight, or because the parameter had a low prevalence within 24 h of birth. 18 candidate variables were selected for inclusion in the modelling process (panel).Figure 1 Flow chart showing filtration of parameters from existing risk scores to select candidate variables

ility in preterm neonates or neonates with low birthweight, or because the parameter had a low prevalence within 24 h of birth. 18 candidate variables were selected for inclusion in the modelling process (panel).Figure 1 Flow chart showing filtration of parameters from existing risk scores to select candidate variables LMICs=low-income and middle-income countries. NNRD=National Neonatal Research Database. Panel Candidate variables evaluated in the modelling process Clinical signs and observations • Heart rate at admission • Respiratory rate at admission • Temperature at admission • Oxygen saturation (SpO2) at admission • Convulsions within 24 h of birth, defined as the presence of any clinical or electrographic seizures • Clinically relevant increase in apnoea or brachycardia episodes, oxygen requirement, ventilatory support, or respiratory rate within 24 h of birth* Therapy-based variables • Bag-mask resuscitation at delivery • Intravenous fluids within 24 h of birth • Antibiotic therapy within 24 h of birth • Oxygen therapy within 24 h of birth† • Highest level of respiratory support administered at any point within 24 h of birth‡ • Caffeine (or aminophylline) within 24 h of birth • Anticonvulsant therapy within 24 h of birth Perinatal factors • Sex • Birthweight • Gestational age • Small for gestational age§ • Presence of visually recognisable anomaly at birth¶ SpO2=peripheral capillary oxygen saturation. FiO2=fraction of inspired oxygen.

• Highest level of respiratory support administered at any point within 24 h of birth‡ • Caffeine (or aminophylline) within 24 h of birth • Anticonvulsant therapy within 24 h of birth Perinatal factors • Sex • Birthweight • Gestational age • Small for gestational age§ • Presence of visually recognisable anomaly at birth¶ SpO2=peripheral capillary oxygen saturation. FiO2=fraction of inspired oxygen. 110 176 neonates were included in the UK development and validation samples. Characteristics of the samples and participants are shown in table 1. More than half (56·6–58·2%) of the neonates had low birthweight (1500–2000 g), 28·0–28·3% had very low birthweight (1000–1499 g), and 13·7–15·1% had extremely low birthweight (<1000 g). Around half (50·4–51·3%) of the neonates were moderate-late preterm (32–36 weeks) and one-third (32·4–33·5%) were very preterm (28–31 weeks). Overall case-fatality was similar across samples (2·8–3·2%). Case-fatality of neonates with extremely low birthweight in the temporal sample (280 [12·5%] of 2238) was lower than in the other samples. No neonatal units declined to contribute data.Table 1 Characteristics of the participants in the data samples from the UK National Neonatal Research Database

oss samples (2·8–3·2%). Case-fatality of neonates with extremely low birthweight in the temporal sample (280 [12·5%] of 2238) was lower than in the other samples. No neonatal units declined to contribute data.Table 1 Characteristics of the participants in the data samples from the UK National Neonatal Research Database Development sample (55 029 eligible neonates; 112 neonatal units) External validation samples Random (40 329 eligible neonates; 75 neonatal units) Temporal (14 818 eligible neonates; 167 neonatal units) Full (55 147 eligible neonates; 173 neonatal units) Birthweight Extremely low birthweight (<1000 g) 7518 (13·7%) 5705 (14·2%) 2238 (15·1%) 7943 (14·4%) Very low birthweight (1000–1499 g) 15 475 (28·1%) 11 290 (28·0%) 4198 (28·3%) 15 488 (28·1%) Low birthweight (1500–2000 g) 32 021 (58·2%) 23 324 (57·9%) 8381 (56·6%) 31 705 (57·5%) Birthweight data missing* 15 (0·03%) 10 (0·02%) 1 (<0·01%) 11 (0·02%) Gestational age (weeks) Extremely preterm (<28) 6969 (12·7%) 5203 (12·9%) 1990 (13·4%) 7193 (13·1%) Very preterm (28–31) 17 810 (32·4%) 13 108 (32·5%) 4963 (33·5%) 18 071 (32·8%) Moderate-late preterm (32–36) 28 241 (51·3%) 20 604 (51·1%) 7470 (50·4%) 28 074 (50·9%) Full term (37–42) 1996 (3·6%) 1408 (3·5%) 393 (2·7%) 1801 (3·3%) Gestational age data missing 13 (0·02%) 6 (0·02%) 2 (0·01%) 8 (0·01%) Size at gestation Small for gestational age 11 039 (20·1%) 7965 (19·8%) 2816 (19·0%) 10 781 (19·6%) Size at gestation data missing 16 (0·03%) 10 (0·03%) 2 (0·01%) 12 (0·02%) Sex Male 27 361 (49·9%) 20 307 (50·4%) 7490 (50·6%) 27 797 (50·4%) Sex data missing 72 (0·1%) 30 (0·07%) 18 (0·1%) 48 (0·09%) Mode of delivery Spontaneous vaginal 16 361 (32·3%) 12 404 (32·6%) 4227 (30·5%) 16 631 (32·0%) Caesarean section 32 473 (64·1%) 24 404 (64·0%) 9148 (66·1%) 33 552 (64·6%) Assisted vaginal 1820 (3·6%) 1284 (3·4%) 463 (3·3%) 1747 (3·4%) Mode of delivery data missing 4375 (8·0%) 2237 (5·5%) 980 (6·6%) 3217 (5·8%) Multiple birth Yes 16 933 (30·8%) 12 056 (29·9%) 4442 (30·0%) 16 498 (29·9%) Multiple birth data missing 22 (0·04%) 8 (0·02%) 2 (0·01%) 10 (0·02%) Location of birth Inborn† 53 954 (98·1%) 39 481 (98·0%) 14 476 (97·9%) 53 957 (98·0%) Location of birth data missing 7 (0·01%) 36 (0·09%) 33 (0·2%) 69 (0·1%) Location of care Neonatal intensive care unit‡ 24 018 (43·7%) 22 362 (55·3%) 7506 (50·7%) 29 840 (54·1%) Local neonatal unit§ 26 276 (47·8%) 13 541 (33·5%) 6054 (40·9%) 19 541 (35·4%) Special care baby unit¶ 4730 (8·6%) 4538 (11·2%) 1258 (8·5%) 5766 (10·5%) Location of care

ing 7 (0·01%) 36 (0·09%) 33 (0·2%) 69 (0·1%) Location of care Neonatal intensive care unit‡ 24 018 (43·7%) 22 362 (55·3%) 7506 (50·7%) 29 840 (54·1%) Local neonatal unit§ 26 276 (47·8%) 13 541 (33·5%) 6054 (40·9%) 19 541 (35·4%) Special care baby unit¶ 4730 (8·6%) 4538 (11·2%) 1258 (8·5%) 5766 (10·5%) Location of care data missing 5 (0·01%) 0 (0·0%) 0 (0·0%) 0 (0·0%) Age at admission (min) Median (IQR) 21 (13–33) 21 (13–34) 23 (15–35) 22 (14–34) Age at admission data missing 5 (0·01%) 0 (0·0%) 0 (0·0%) 0 (0·0%) Disposition Died before discharge 1653 (3·0%) 1306 (3·2%) 395 (2·8%) 1701 (3·1%) Extremely low birthweight (<1000 g) 1159 (15·4%) 929 (16·3%) 280 (12·5%) 1209 (15·2%) Very low birthweight (1000–1499 g) 295 (1·9%) 228 (2·0%) 67 (1·6%) 295 (1·9%) Low birthweight (1500–2000 g) 199 (0·6%) 149 (0·6%) 48 (0·6%) 197 (0·6%) Died within 24 h of birth 207 (0·4%) 194 (0·5%) 50 (0·3%) 244 (0·4%) Transferred to another care unit‖ 12 793 (23·3%) 10 119 (25·1%) 4268 (30·3%) 14387 (26·5%) Disposition data missing 73 (0·1%) 32 (0·1%) 726 (4·9%) 758 (1·4%) Age at discharge (days) Median (IQR) 22 (12–38) 21 (11–36) 19 (10–34) 20 (11–36) Age at discharge data missing 21 (0·04%) 13 (0·03%) 740 (5·0%) 754 (1·4%) Variables collected at time of birth Bag-mask resuscitation at delivery 24 302 (44·2%) 18 297 (45·4%) 6131 (41·4%) 24 428 (44·3%) Visually recognisable anomaly 1299 (2·4%) 1151 (2·9%) 309 (2·1%) 1460 (2·7%) Variables collected at time of admission Heart rate (beats per min), mean (SD) 153·4 (18·4) 153·8 (18·7) 154·5 (18·6) 154·0 (18·6) Heart rate data missing 7197 (13·1%) 3557 (8·8%) 1234 (8·3%) 4791 (8·7%) Respiratory rate (breaths per min), mean (SD) 53·3 (29·8) 52·4 (12·9) 52·5 (13·2) 52·4 (13·0) Respiratory rate data missing 9535 (17·3%) 5377 (13·3%) 2008 (13·6%) 7385 (13·4%) Temperature (°C), mean (SD) 36·6 (0·7) 36·6 (0·7) 36·7 (0·8) 36·6 (0·7) Temperature data missing 589 (1·1%) 371 (0·9%) 166 (1·1%) 537 (1·0%) SpO2 (%), median (IQR) 96 (93–98) 96 (92–99) 96 (93–99) 96 (93–99) SpO2 data missing 7787 (14·2%) 4213 (10·4%) 1392 (9·4%) 5605 (10·2%) Variables collected within 24 h of birth Increased apnoea or bradycardia, oxygen, ventilatory support, or respiratory rate 3005 (5·5%) 2458 (6·1%) 903 (6·1%) 3361 (6·1%) Convulsions 134 (0·3%) 81 (0·2%) 24 (0·2%) 105 (0·2%) Convulsions data missing 579 (1·1%) 456 (1·1%) 297 (2·0%) 753 (1·4%) Oxygen therapy 13 998 (25·4%) 10 989 (27·3%) 5178 (34·9%) 16 167 (29·3%) Highest level of respiratory support Nasal cannula or headbox 3722 (7·0%)

ate 3005 (5·5%) 2458 (6·1%) 903 (6·1%) 3361 (6·1%) Convulsions 134 (0·3%) 81 (0·2%) 24 (0·2%) 105 (0·2%) Convulsions data missing 579 (1·1%) 456 (1·1%) 297 (2·0%) 753 (1·4%) Oxygen therapy 13 998 (25·4%) 10 989 (27·3%) 5178 (34·9%) 16 167 (29·3%) Highest level of respiratory support Nasal cannula or headbox 3722 (7·0%) 2758 (7·0%) 2035 (13·9%) 4793 (8.9%) CPAP, BiPAP or SiPAP, or invasive ventilation 23 374 (43·8%) 16 676 (42·6%) 6427 (43·8%) 23 103 (42·9%) Respiratory support data missing 1658 (3·0%) 1161 (2·9%) 150 (1·0%) 1311 (2·4%) Other interventions Intravenous fluids 41 506 (75·4%) 30 468 (75·6%) 11 697 (78·9%) 42 165 (76·5%) Antibiotic therapy 39 774 (72·3%) 29 877 (74·1%) 11 152 (75·3%) 41 029 (74·4%) Caffeine citrate 14 276 (25·9%) 10 862 (26·9%) 5438 (36·7%) 16 300 (29·6%) Anticonvulsant therapy 162 (0·3%) 150 (0·4%) 45 (0·3%) 195 (0·4%) Data are complete except where missing data are detailed; missing data are the total number of neonates for whom data are not available. Data are n (%) except where otherwise indicated. See panel for definitions of variables. CPAP=continuous positive airway pressure. BiPAP=bilevel positive airway pressure. SiPAP=synchronised intermittent positive airway pressure. SpO2=peripheral capillary oxygen saturation. * For neonates whose birthweight was missing, admission weight was used to determine eligibility. † Inborn is defined as birth at the hospital of neonatal unit admission.

Data are n (%) except where otherwise indicated. See panel for definitions of variables. CPAP=continuous positive airway pressure. BiPAP=bilevel positive airway pressure. SiPAP=synchronised intermittent positive airway pressure. SpO2=peripheral capillary oxygen saturation. * For neonates whose birthweight was missing, admission weight was used to determine eligibility. † Inborn is defined as birth at the hospital of neonatal unit admission. ‡ Intensive care units provide care for the sickest neonates who require constant supervision and monitoring, including those born at fewer than 27 weeks’ gestational age: care typically includes mechanical ventilation; surgery services offered in some units; care is analogous to American Academy of Paediatrics levels 3 and 4.27 § Local neonatal units provide full care for the majority of babies more than 27 weeks’ gestational age, including short periods of intensive care; therapies provided include continuous monitoring, CPAP, and parenteral nutrition. ¶ Special care units provide care for all other babies who could not reasonably be cared for at home; therapies provided include cardiorespiratory monitoring, nasogastric feeding, supplemental oxygen, and phototherapy. ‖ Transfer to another care unit from the initial unit of neonatal admission.

§ Local neonatal units provide full care for the majority of babies more than 27 weeks’ gestational age, including short periods of intensive care; therapies provided include continuous monitoring, CPAP, and parenteral nutrition. ¶ Special care units provide care for all other babies who could not reasonably be cared for at home; therapies provided include cardiorespiratory monitoring, nasogastric feeding, supplemental oxygen, and phototherapy. ‖ Transfer to another care unit from the initial unit of neonatal admission. Characteristics of the 550 neonates in the Gambian validation sample are shown in table 2. Among the 550 neonates, 298 (54·2%) had a low birthweight, 189 (34·4%) had a very low birthweight, and 63 (11·5%) had an extremely low birthweight. 142 (25·8%) of 550 neonates were multiple births (eg, twins), 299 (54·5%) of 549 were inborn, and 215 (41·4%) of 520 died.Table 2 Characteristics of participants in the Gambian validation sample

, 298 (54·2%) had a low birthweight, 189 (34·4%) had a very low birthweight, and 63 (11·5%) had an extremely low birthweight. 142 (25·8%) of 550 neonates were multiple births (eg, twins), 299 (54·5%) of 549 were inborn, and 215 (41·4%) of 520 died.Table 2 Characteristics of participants in the Gambian validation sample Neonates with available data* Neonates with characteristic Birthweight† 550 (100%) .. Extremely low birthweight (<1000 g) .. 63 (11·5%) Very low birthweight (1000–1499 g) .. 189 (34·4%) Low birthweight (1500 to 2000 g) .. 298 (54·2%) Sex† 549 (99·8%) .. Male sex .. 261 (47·5%) Mode of delivery† 488 (88·7%) .. Spontaneous vaginal .. 342 (70·1%) Caesarean section .. 140 (28·7%) Assisted vaginal .. 6 (1·2%) Multiple birth† 550 (100%) .. Yes .. 142 (25·8%) Inborn‡† 549 (99·8%) .. Yes .. 299 (54·5%) Died before discharge† 520 (94·5%) .. Total .. 215 (41·4%) Extremely low birthweight (<1000 g) .. 55/61 (90·2%)§ Very low birthweight (1000–1499 g) .. 93/179 (52·0%)§ Low birthweight (1500–2000 g) .. 67/280 (23·9%)§ Oxygen saturation at admission 513 (93·3%) .. SpO2 (%) at admission‖ .. 92% (83–96) Highest level of respiratory support within 24 h of birth† 494 (89·8%) .. Nasal cannula .. 294 (59·5%) CPAP ventilation .. 53 (10·7%) Data are n (%), n/N (%), median (IQR). See panel for definition of variables. CPAP=continuous positive airway pressure. SpO2=peripheral capillary oxygen saturation. * Out of the total 550 neonates. † Information is included on routine admission forms. ‡ Defined as birth at the study hospital.

Neonates with available data* Neonates with characteristic Birthweight† 550 (100%) .. Extremely low birthweight (<1000 g) .. 63 (11·5%) Very low birthweight (1000–1499 g) .. 189 (34·4%) Low birthweight (1500 to 2000 g) .. 298 (54·2%) Sex† 549 (99·8%) .. Male sex .. 261 (47·5%) Mode of delivery† 488 (88·7%) .. Spontaneous vaginal .. 342 (70·1%) Caesarean section .. 140 (28·7%) Assisted vaginal .. 6 (1·2%) Multiple birth† 550 (100%) .. Yes .. 142 (25·8%) Inborn‡† 549 (99·8%) .. Yes .. 299 (54·5%) Died before discharge† 520 (94·5%) .. Total .. 215 (41·4%) Extremely low birthweight (<1000 g) .. 55/61 (90·2%)§ Very low birthweight (1000–1499 g) .. 93/179 (52·0%)§ Low birthweight (1500–2000 g) .. 67/280 (23·9%)§ Oxygen saturation at admission 513 (93·3%) .. SpO2 (%) at admission‖ .. 92% (83–96) Highest level of respiratory support within 24 h of birth† 494 (89·8%) .. Nasal cannula .. 294 (59·5%) CPAP ventilation .. 53 (10·7%) Data are n (%), n/N (%), median (IQR). See panel for definition of variables. CPAP=continuous positive airway pressure. SpO2=peripheral capillary oxygen saturation. * Out of the total 550 neonates. † Information is included on routine admission forms. ‡ Defined as birth at the study hospital. § Proportion of babies in each birthweight category who died (outcome data were not available for 5·5% of babies). ‖ Information collected for the trial (eKMC trial).

Neonates with available data* Neonates with characteristic Birthweight† 550 (100%) .. Extremely low birthweight (<1000 g) .. 63 (11·5%) Very low birthweight (1000–1499 g) .. 189 (34·4%) Low birthweight (1500 to 2000 g) .. 298 (54·2%) Sex† 549 (99·8%) .. Male sex .. 261 (47·5%) Mode of delivery† 488 (88·7%) .. Spontaneous vaginal .. 342 (70·1%) Caesarean section .. 140 (28·7%) Assisted vaginal .. 6 (1·2%) Multiple birth† 550 (100%) .. Yes .. 142 (25·8%) Inborn‡† 549 (99·8%) .. Yes .. 299 (54·5%) Died before discharge† 520 (94·5%) .. Total .. 215 (41·4%) Extremely low birthweight (<1000 g) .. 55/61 (90·2%)§ Very low birthweight (1000–1499 g) .. 93/179 (52·0%)§ Low birthweight (1500–2000 g) .. 67/280 (23·9%)§ Oxygen saturation at admission 513 (93·3%) .. SpO2 (%) at admission‖ .. 92% (83–96) Highest level of respiratory support within 24 h of birth† 494 (89·8%) .. Nasal cannula .. 294 (59·5%) CPAP ventilation .. 53 (10·7%) Data are n (%), n/N (%), median (IQR). See panel for definition of variables. CPAP=continuous positive airway pressure. SpO2=peripheral capillary oxygen saturation. * Out of the total 550 neonates. † Information is included on routine admission forms. ‡ Defined as birth at the study hospital. § Proportion of babies in each birthweight category who died (outcome data were not available for 5·5% of babies). ‖ Information collected for the trial (eKMC trial). The full model (18 variables) had a c-index of 0·9223 in the development sample (n=41 514). After stepwise elimination, the final model included three variables (table 3), with a c-index of 0·8883 and a Brier score of 0·0232 (table 4). Complete data on all three variables were available for 46 108 (83·8%) of 55 029 neonates in the development sample. After imputation of missing values for predictor variables (n=54 956), the resulting β coefficients were nearly identical to original estimates (appendix p 9) and model performance was unchanged (c-index 0·8894; appendix p 2). Admission age was uncertain for 5 (0·01%) of 55 029 neonates; in a sensitivity analysis excluding these neonates, there was no change in performance (c-index 0·8886, Brier score 0·0232). 12 793 (23·3%) of 54 956 neonates were transferred from the unit of admission to another care unit; an analysis excluding these neonates showed improved performance (c-index 0·9150, Brier score 0·0255; appendix p 2). Predictive accuracy for mortality within 24 h (c-index 0·8858, Brier score 0·0037) was nearly identical to that for in-hospital mortality. Because availability of SpO2 monitoring is variable in LMIC settings, we tested a related variable (clinically relevant increase in apnoea or bradycardia, oxygen requirement, ventilatory support, or respiratory rate); however, this variable was not associated with in-hospital mortality (c-index 0·5061). A plot of observed versus predicted mortality risk in the development sample is shown in figure 2.Table 3 Derivation logistic model for the NMR-2000 score

or bradycardia, oxygen requirement, ventilatory support, or respiratory rate); however, this variable was not associated with in-hospital mortality (c-index 0·5061). A plot of observed versus predicted mortality risk in the development sample is shown in figure 2.Table 3 Derivation logistic model for the NMR-2000 score β coefficient 95% confidence interval* Integer-points† c-index Birthweight (g) −0·0032 −0·0035 to −0·0029 Birthweight/100 0·8540 Highest respiratory support within first 24 h .. .. .. 0·7529 Nasal cannula or headbox 0·3167 −0·1055 to 0·7389‡ −1 .. CPAP, BiPAP or SiPAP, or invasive ventilation 1·6214 1·2682 to 1·9746‡ −5 .. SpO2 at admission −0·0390 −0·0455 to −0·0326 .. 0·6712 <80% (reference level) .. .. 0§ .. 80–89% −0·7694 −1·0093 to −0·5294 2§ .. 90–100% −1·3697 −1·6019 to −1·1376 4§ .. Constant 2·6142¶ 1·7655 to 3·4629 .. .. n=46 108. CPAP=continuous positive airway pressure. BiPAP=bilevel positive airway pressure. SiPAP=synchronised intermittent positive airway pressure. SpO2=peripheral capillary oxygen saturation. * p<0·0001 for estimates for all variables. † Calculated by multiplying the β coefficient by a constant (–3·13) and rounding to the nearest integer. The reciprocal of the coefficient for birthweight divided by 100 ([1/–0·0032]/100=–3·13) was used as the constant to retain the exact birthweight (per 100 g) in the score. ‡ p<0·0001 for overall effect of level of respiratory support.

† Calculated by multiplying the β coefficient by a constant (–3·13) and rounding to the nearest integer. The reciprocal of the coefficient for birthweight divided by 100 ([1/–0·0032]/100=–3·13) was used as the constant to retain the exact birthweight (per 100 g) in the score. ‡ p<0·0001 for overall effect of level of respiratory support. § The continuous SpO2 parameter was categorised into clinically meaningful categorical variables; the β coefficients of these variables were multiplied by the constant to obtain integer-points; the reference level (<80%) was assigned zero points. ¶ Reflects β coefficient for constant in model including SpO2 as a continuous variable. Table 4 Model performance in the development and validation samples Development sample (n=46 108) External validation samples Original Optimism-adjusted* Random (n=35 193) Temporal (n=12 653) Full (n=47 846) Gambian (n=457) Brier score 0·0232 0·0233 0·0271 0·0240 0·0263 0·1688 c-index 0·8883 0·8882 0·8930 0·8859 0·8912 0·8170 * Because optimism-adjusted estimates of the c-index and Brier score were nearly identical to the original estimates, no adjustments were made to the model coefficients. Figure 2 Predicted versus observed death for population deciles by predicted risk in the development sample n=46 108. Graph created using pmcalplot.

Development sample (n=46 108) External validation samples Original Optimism-adjusted* Random (n=35 193) Temporal (n=12 653) Full (n=47 846) Gambian (n=457) Brier score 0·0232 0·0233 0·0271 0·0240 0·0263 0·1688 c-index 0·8883 0·8882 0·8930 0·8859 0·8912 0·8170 * Because optimism-adjusted estimates of the c-index and Brier score were nearly identical to the original estimates, no adjustments were made to the model coefficients. Figure 2 Predicted versus observed death for population deciles by predicted risk in the development sample n=46 108. Graph created using pmcalplot. Birthweight was the most predictive variable in the model (c-index 0·8540). The reciprocal of the coefficient for birthweight divided by 100 ([1/–0·0032]/100=–3·13) was used as the constant to enable retention of exact birthweights in score calculation (table 3),28 thereby improving predictive ability. The score range for low risk was set at 16 or more, for medium risk at 6–15, and for high risk at 5 or fewer points (appendix p 9). An example score form is shown in figure 3. Among 46 108 neonates from the development sample with complete data on the three variables included in the final model, 27 289 (59·1%) were designated as low risk, 17 215 (37·3%) as medium risk, and 1640 (3·6%) as high risk. Observed risks were 0·3% (95% CI 0·3–0·4) for low risk, 4·1% (3·8–4·4) for medium risk, and 27·3% (25·2–29·5) for high risk, with a c-index of 0·8875. Mean predicted risks derived from regression coefficients were 0·2% (SD 0·2) for low risk, 4·6% (SD 4·2) for medium risk, and 23·5% (SD 8·8) for high risk. Observed risks across population deciles by score were similar to the risks predicted with regression coefficients (appendix p 10).Figure 3 NMR-2000, a simplified risk score to predict mortality amongst neonates weighing 2000 g or less

0·2) for low risk, 4·6% (SD 4·2) for medium risk, and 23·5% (SD 8·8) for high risk. Observed risks across population deciles by score were similar to the risks predicted with regression coefficients (appendix p 10).Figure 3 NMR-2000, a simplified risk score to predict mortality amongst neonates weighing 2000 g or less (A) An example mortality risk score for clinical use. (B) Predicted risk of in-hospital mortality plotted against risk scores in the development sample; bar data indicate proportion of neonates; blue line indicates predicted mortality risk. CPAP=continuous positive airway pressure. BiPAP=bilevel positive airway pressure. SiPAP=synchronised intermittent positive airway pressure.

edicted risk of in-hospital mortality plotted against risk scores in the development sample; bar data indicate proportion of neonates; blue line indicates predicted mortality risk. CPAP=continuous positive airway pressure. BiPAP=bilevel positive airway pressure. SiPAP=synchronised intermittent positive airway pressure. After bootstrap resampling, optimism-adjusted estimates of c-index and Brier score were nearly identical to the original measures; thus, no adjustments were made to the coefficients. In the random validation sample, complete data on all three parameters were available for 35 193 (87·3%) of 40 329 neonates, for the temporal validation sample the data were available for 12 653 (85·4%) of 14 818 neonates, for the full validation sample they were available for 47 846 (86·8%) of 55 147 neonates, and for the Gambian validation sample complete data on all three parameters were available for 457 (83·1%) of 550 neonates. The model showed very good performance across the UK validation samples (c-index 0·8859–0·8930) and good performance in the Gambian validation sample (c-index 0·8170; Brier score 0·1688; table 4). Performance was similar among neonates weighing 1500 g or less in the Gambian sample (c-index 0·8069, Brier score 0·1753).

The model showed very good performance across the UK validation samples (c-index 0·8859–0·8930) and good performance in the Gambian validation sample (c-index 0·8170; Brier score 0·1688; table 4). Performance was similar among neonates weighing 1500 g or less in the Gambian sample (c-index 0·8069, Brier score 0·1753). Graphical plots showed a high level of agreement between observed and predicted mortality risk across the external validation samples (figure 4). Applying the empirical optimal cutpoint of 3·9% based on the Youden Index gave moderately high sensitivity (79·1–81·6) and specificity (81·0–82·9), with high negative predictive value (99·3) and low positive predictive value (11·4–12·2; appendix p 10).Figure 4 Predicted versus observed mortality risk for population deciles in the random, temporal, full, and Gambian validation samples Graphs created using pmcalplot.

Graphical plots showed a high level of agreement between observed and predicted mortality risk across the external validation samples (figure 4). Applying the empirical optimal cutpoint of 3·9% based on the Youden Index gave moderately high sensitivity (79·1–81·6) and specificity (81·0–82·9), with high negative predictive value (99·3) and low positive predictive value (11·4–12·2; appendix p 10).Figure 4 Predicted versus observed mortality risk for population deciles in the random, temporal, full, and Gambian validation samples Graphs created using pmcalplot. The discriminatory ability of the simplified integer score was similar to the model using regression coefficients, with c-indices of 0·8903 in the full validation sample and 0·8082 in the Gambian validation sample (appendix p 11). Among the 47 846 neonates in the UK full validation sample, 28 565 (59·7%) were designated as low risk, 17 407 (36·4%) as medium risk, and 1874 (3·9%) as high risk. Observed risks for these categories were 0·4% (95% CI 0·3–0·5) for low risk, 4·8% (4·5–5·1) for medium risk, and 29·7% (27·7–31·8) for high risk. In the Gambian validation sample, the score range for low risk was set at 23 or more, for medium risk at 17–22, and for high risk at 16 or fewer points (appendix pp 3, 11). Among the 457 neonates in the Gambian sample for whom data on all three parameters were available, 28 (6·1%) were designated as low risk, 215 (47·1%) as medium risk, and 214 (46·8%) as high risk. Observed risks were 10·7% (95% CI 3·5–28·5) for low risk, 21·4% (16·4–27·4) for medium risk, and 68·2% (61·7–74·1) for high risk. Mean predicted risks derived from regression coefficients were 9·4% (SD 1·9) for low risk, 22·3% (SD 8·5) for medium risk, and 67·4% (SD 18·4) for high risk. Observed risks across population deciles by score were similar to those predicted with coefficients (appendix p 11).

um risk, and 68·2% (61·7–74·1) for high risk. Mean predicted risks derived from regression coefficients were 9·4% (SD 1·9) for low risk, 22·3% (SD 8·5) for medium risk, and 67·4% (SD 18·4) for high risk. Observed risks across population deciles by score were similar to those predicted with coefficients (appendix p 11). Comparison of areas under the ROC curves for NMR-2000 (c-index 0·8523 [95% CI 0·8336–0·8710]) and CRIB-II (c-index 0·7443 [95% CI 0·7153–0·7733]) among 10 812 neonates born at 32 weeks' gestation or earlier (figure 5) indicated that discriminatory performance of NMR-2000 was superior to that of CRIB-II (p<0·0001).Figure 5 Comparison of areas under the ROC curves for CRIB-II and NMR-2000 This analysis includes neonates born at 32 weeks' gestation or earlier in the full validation sample (n=10 812). ROC=receiver operating characteristic. Discussion This population-wide study, including data from 110 176 newborn babies at 187 hospitals in the UK and 550 newborn babies at one hospital in The Gambia, has derived and validated NMR-2000 for predicting in-hospital mortality. A strength of this work is that, to our knowledge, this is the largest dataset that has been used to develop and validate a neonatal mortality risk score. Among neonates born at 32 weeks' gestation or earlier, the discriminatory ability of NMR-2000 was superior to that of CRIB-II, one of the most widely used neonatal risk scores.

of this work is that, to our knowledge, this is the largest dataset that has been used to develop and validate a neonatal mortality risk score. Among neonates born at 32 weeks' gestation or earlier, the discriminatory ability of NMR-2000 was superior to that of CRIB-II, one of the most widely used neonatal risk scores. Performance of the NMR-2000 simplified integer score, which can be measured and calculated at the bedside, was similar to that of the model using regression coefficients. The three parameters used in the score can be feasibly collected in LMIC settings. Although sub-Saharan Africa and southern Asia account for 78% of the world's neonatal deaths,1 existing risk scores have primarily been developed for intensive care settings and often require complex calculations. In LMICs, where parameters in widely used scores are typically not available nor reliably measurable,20, 21 NMR-2000 could support shared decision making by enabling providers to objectively assess illness severity.29 The score could be used in clinical trials to assess eligibility and compare participants.29 Additionally, NMR-2000 could inform service delivery planning by identifying bottlenecks in care provision.6 Given that 73% of neonatal deaths occur within the first 7 days of life,24 early recognition of severe illness and rapid initiation of evidence-based interventions are crucial to promoting survival.5, 9, 12

s.29 Additionally, NMR-2000 could inform service delivery planning by identifying bottlenecks in care provision.6 Given that 73% of neonatal deaths occur within the first 7 days of life,24 early recognition of severe illness and rapid initiation of evidence-based interventions are crucial to promoting survival.5, 9, 12 The c-index was 0·8859–0·8930 across the development and UK validation samples, suggesting that NMR-2000 can discriminate neonates who will die from neonates who will survive. This level of performance is similar to that of commonly used neonatal mortality scores in high-resource settings. Discriminatory ability at the time of model derivation ranged from 0·87 for TRIPS-II16 to 0·92 for CRIB-II.18 Similar to TRIPS-II, NMR-2000 can be assessed at any point within the first 24 h of life and could be repeated if the level of respiratory support increases. The performance of NMR-2000 (c-index 0·8523) was superior to that of CRIB-II (0·7443) among neonates born at 32 weeks' gestation or earlier. The model also showed very good predictive ability for mortality within 24 h of birth (c-index 0·8858), which is notable because 37% of neonatal deaths in sub-Saharan Africa occur within this timeframe.24 The NMR-2000 model showed a high level of agreement between observed and predicted deaths, as assessed by calibration plots, in the development and validation samples. Calibration plots are the preferred method for assessing calibration.25 Previous neonatal scores, including NTISS,13 SNAPPE-II,15 CRIB-II,18 and TRIPS-II,16 were reported to have good calibration for predicting in-hospital mortality using the Hosmer-Lemeshow test. However, such results should be interpreted with caution given the limitations of this test, which include subjective and imprecise grouping of babies as well as inability to denote the directionality of miscalibration when incongruities are detected.25

r predicting in-hospital mortality using the Hosmer-Lemeshow test. However, such results should be interpreted with caution given the limitations of this test, which include subjective and imprecise grouping of babies as well as inability to denote the directionality of miscalibration when incongruities are detected.25 In the Gambian validation sample, the NMR-2000 model had good discrimination and overall goodness-of-fit, with c-index of 0·8170 and Brier score of 0·1688. Complete data were available for 83% of neonates. The calibration plot showed a strong agreement between observed and predicted mortality. These findings suggest that NMR-2000 is valid for use in LMIC settings where pulse oximetry is available. Discrimination of the SAWS score, developed for neonates weighing 1500 g or less in low-resource settings, at the time of validation (c-index 0·679–0·698)22 was decreased relative to NMR-2000 among Gambian neonates weighing 1500 g or less (c-index 0·8069). Notably, neither goodness-of-fit nor calibration were reported for the SAWS score.22 Further, SAWS relies on accurate assessment of gestational age, which can be challenging in LMICs because of late presentation for antenatal care, poor recall of last menstrual period, and unavailability of ultrasonography.30 Case-fatality of Gambian neonates in this study is similar to that reported from a previous study at EFSTH (35% overall, 58% for neonates with a very low birthweight),23 and higher than studies at similar hospitals in Ghana (20% overall),31 Nigeria (14–20% overall),32, 33 and Burkina Faso (15% overall).34

sonography.30 Case-fatality of Gambian neonates in this study is similar to that reported from a previous study at EFSTH (35% overall, 58% for neonates with a very low birthweight),23 and higher than studies at similar hospitals in Ghana (20% overall),31 Nigeria (14–20% overall),32, 33 and Burkina Faso (15% overall).34 Among the three NMR-2000 parameters, all except SpO2 are included on routine admission forms at EFSTH, Gambia (at time of screening),23 as well as standard forms at government hospitals in Kenya, Malawi, and Tanzania. We were able to obtain SpO2 data for the Gambian sample primarily because these data were being collected as part of the eKMC trial screening process. Variability in the implementation of routine pulse oximetry is a crucial gap in low-resource neonatal units.35, 36 In a study of nearly 7500 neonates admitted to 11 hospitals in Nigeria, hypoxaemia increased the adjusted odds of mortality by six times, and clinical signs (eg, chest in-drawing, grunting) poorly predicted hypoxaemia.37 Furthermore, expansion of neonatal inpatient care, often of variable quality and frequently inclusive of unmonitored 100% oxygen supplementation, has placed sub-Saharan Africa on the brink of an epidemic of retinopathy of prematurity.38 Widespread availability of SpO2 monitoring and improved coverage of screening for and treatment of retinopathy of prematurity will be essential to control the incidence of visual loss in affected neonates. In LMIC settings, successful implementation of NMR-2000 would require sensitisation around recording the three parameters. Several studies have highlighted issues surrounding the collection of data on neonatal care in LMICs, including variable uptake of standard admission forms;36 incomplete documentation of assessments, monitoring, and therapies prescribed;20, 23, 36 and low capacity of data systems to capture information on neonates who die soon after birth or are transferred to another facility.4 Increasing the quality and coverage of data is crucial to promote actions to improve neonatal survival, and will require coordination across different levels of the health-care system.

6 and low capacity of data systems to capture information on neonates who die soon after birth or are transferred to another facility.4 Increasing the quality and coverage of data is crucial to promote actions to improve neonatal survival, and will require coordination across different levels of the health-care system. One strength of this study is our use of a large and purposely-selected UK dataset, which enabled maximisation of model performance. One limitation is that the Gambian dataset was small and limited to a single hospital; research is required to validate the model using a larger LMIC dataset. Several candidate variables in the development sample had a considerable proportion of missing data, including admission heart rate, respiratory rate, and SpO2. It was not possible to compare NMR-2000 with the CRIB, SNAP, SNAP-II, SNAPPE-II, or TRIPS-II scores, because the NNRD did not include all parameters required for their calculation. Because pulse oximetry is not always available in low-resource neonatal units, the usefulness of the NMR-2000 score in such settings could be limited. The NNRD did not include clinical signs of respiratory distress that could be tested as a potential proxy for SpO2. We tested a related variable (clinically relevant increase in apnoea or bradycardia, oxygen requirement, ventilatory support, or respiratory rate); however, this variable was not associated with mortality. The use of respiratory support level as a parameter could affect the performance of the model. Administration of therapies varies in line with variations in clinical practice19 and resource availability and so might not reflect true therapeutic requirements.36 In LMICs, delivery systems for oxygen therapy and CPAP might be unavailable or non-functional, and related supplies (eg, nasal cannulas) might be out of stock.8, 35, 36

of therapies varies in line with variations in clinical practice19 and resource availability and so might not reflect true therapeutic requirements.36 In LMICs, delivery systems for oxygen therapy and CPAP might be unavailable or non-functional, and related supplies (eg, nasal cannulas) might be out of stock.8, 35, 36 Research is required to validate the NMR-2000 score in low-resource settings using a sufficiently sized dataset, and to evaluate its usefulness for supporting clinical decision making.29 A follow-up study using a large, multihospital dataset from Kenya is planned. Nurses have essential roles as frontline providers of neonatal care; however, there is a severe shortage of neonatal nurses in LMICs.6, 7 Future research could explore the model's ability to inform resource use,13 particularly nursing workload. The NMR-2000 is a simplified risk score, validated for high-resource and low-resource settings where pulse oximetry is available, to accurately predict in-hospital mortality among neonates weighing 2000 g or less. By enabling providers to objectively assess illness severity, this tool could contribute to improvements in the quality of care delivered in LMIC facilities. Early recognition of severe illness and rapid initiation of evidence-based interventions are crucial to promoting survival of small and vulnerable neonates. Data sharing Data from the UK National Neonatal Research Database is available upon request. Supplementary Material Supplementary appendix

The NMR-2000 is a simplified risk score, validated for high-resource and low-resource settings where pulse oximetry is available, to accurately predict in-hospital mortality among neonates weighing 2000 g or less. By enabling providers to objectively assess illness severity, this tool could contribute to improvements in the quality of care delivered in LMIC facilities. Early recognition of severe illness and rapid initiation of evidence-based interventions are crucial to promoting survival of small and vulnerable neonates. Data sharing Data from the UK National Neonatal Research Database is available upon request. Supplementary Material Supplementary appendix Acknowledgments This work was supported by a grant from the Bill & Melinda Gates Foundation (OPP1107312) to the University of California San Francisco Preterm Birth Initiative. Funds from the Eunice Kennedy Shriver National Institute of Child Health and Human Development of the National Institutes of Health (K23HD092611) awarded to MMM, the Wellcome Trust (2000116) awarded to HB, and the Joint Global Health Trials scheme of the Department of Health and Social Care, the Department for International Development, the Medical Research Council, and the Wellcome Trust (MR/S004971/1) awarded to JEL also supported this study. The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health. We acknowledge the invaluable assistance of Richard Colquhoun and Kayleigh Ougham at the Neonatal Data Analysis Unit (Imperial College London, London, UK), Swetlana Kapoor from the American University of The Gambia (Serrekunda, The Gambia), and the clinical teams from all contributing neonatal units who recorded data. We are grateful to all the families who agreed to the inclusion of their infants' data in the UK National Neonatal Research Database.

e London, London, UK), Swetlana Kapoor from the American University of The Gambia (Serrekunda, The Gambia), and the clinical teams from all contributing neonatal units who recorded data. We are grateful to all the families who agreed to the inclusion of their infants' data in the UK National Neonatal Research Database. Contributors MMM, CT, DE, JEL, and EA conceived the study. MMM, HB, CT, DE, JEL, and EA contributed to study design. MMM and EA developed the analysis plan. MMM, CT, and CG were involved in curation of data from the NNRD. HB and AG collected data in The Gambia. MMM analysed the data and wrote the manuscript. MMM, HB, CT, CG, PW, DE, JEL, and EA interpreted the data. All authors participated in manuscript revision and approval of the final version. Declaration of interests CG has received grants from the Medical Research Council during the conduct of the study; grants from the National Institute for Health Research, Mason Medical Research Foundation, Chiesi Pharmaceuticals, Rosetrees Foundation, and the Canadian Institute for Health Research, outside the submitted work; and personal fees from Chiesi Pharmaceuticals, outside the submitted work. CG is also a voluntary, unremunerated member of the Neonatal Data Analysis Unit Steering Board, which oversees the UK National Neonatal Research Database. JEL is a member of the International Advisory Board for The Lancet Child and Adolescent Health. All other authors declare no competing interests.

the submitted work. CG is also a voluntary, unremunerated member of the Neonatal Data Analysis Unit Steering Board, which oversees the UK National Neonatal Research Database. JEL is a member of the International Advisory Board for The Lancet Child and Adolescent Health. All other authors declare no competing interests. * Defined as an increase that was clinically significant enough to necessitate obtaining a culture to evaluate for suspected sepsis, at any point within 24 h of birth. † Defined as delivery of supplemental oxygen (FiO2 >0·21) via any method at any point within 24 h of birth. ‡ Not including initial resuscitation at birth; level 1 defined as nasal cannula or headbox; level 2 defined as continuous positive airway pressure, bilevel or synchronised intermittent positive airway pressure, or invasive ventilation with an endotracheal tube or tracheostomy. § Defined as birthweight less than the 5th percentile for gestational age, using UK-WHO standards.26 ¶ Defined as the presence of one or more of the following: cleft lip or palate; microcephaly; trisomy 13, trisomy 18, or trisomy 21; spina bifida, myelomeningocele, or meningocele; encephalocele; anencephaly; holoprosencephaly or prosencephaly; ambiguous genitalia; hypospadias; absent anus; gastroschisis; exomphalos or omphalocele; achondroplasia; Noonan syndrome.